ADAPTIVE PARAMETER TRANSFER FOR LEARNING MODELS
A process includes obtaining a directed graph of a symbolic artificial intelligence model used by a first entity, the directed graph comprising a first set of vertices and a first set of edges associating pairs of vertices of the first set of vertices. The method also includes determining a set of features based on the directed graph that includes an identifier of a graph portion template, where each respective vertex of the graph portion template of the graph portion template is labeled with a same category from the set of mutually-exclusive categories as a corresponding respective vertex of a graph portion of the directed graph. The method also includes obtaining a set of model parameter values for a machine learning model based on the graph portion template and providing the set of model parameter values and the graph portion templates the first entity.
This patent claims the benefit of U.S. Provisional Patent Application 62/897,240, filed 6 Sep. 2019, titled “SMART DEONTIC DATA SYSTEMS.” This patent also claims the benefit of U.S. Provisional Patent Application 62/959,418, filed 10 Jan. 2020, titled “GRAPH-MANIPULATION BASED DOMAIN-SPECIFIC ENVIRONMENT.” This patent also claims the benefit of U.S. Provisional Patent Application 62/959,481, filed 10 Jan. 2020, titled “GRAPH OUTCOME DETERMINATION IN DOMAIN-SPECIFIC EXECUTION ENVIRONMENT.” This patent also claims the benefit of U.S. Provisional Patent Application 62/959,377, filed 10 Jan. 2020, titled “SMART DEONTIC MODEL AND SYSTEMS.” This patent also claims the benefit of U.S. Provisional Patent Application 63/020,808, filed 6 May 2020, titled “GRAPH EXPANSION AND OUTCOME DETERMINATION FOR GRAPH-DEFINED PROGRAM STATES.” This patent also claims the benefit of U.S. Provisional Patent Application 63/033,063, filed 1 Jun. 2020, titled “MODIFICATION OF IN-EXECUTION SMART CONTRACT PROGRAMS.” This patent also claims the benefit of U.S. Provisional Patent Application 63/034,255, filed 3 Jun. 2020, titled “SEMANTIC CONTRACT MAPS.” This patent also claims the benefit of U.S. patent application Ser. No. 16/893,290, filed 4 Jun. 2020, titled “GRAPH-MANIPULATION BASED DOMAIN-SPECIFIC EXECUTION ENVIRONMENT.” This patent also claims the benefit of U.S. patent application Ser. No. 16/893,318, filed 4 Jun. 2020, titled “GRAPH OUTCOME DETERMINATION IN DOMAIN-SPECIFIC EXECUTION ENVIRONMENT.” This patent also claims the benefit of U.S. patent application Ser. No. 16/893,295, filed 4 Jun. 2020, titled “MODIFICATION OF IN-EXECUTION SMART CONTRACT PROGRAMS.” This patent also claims the benefit of U.S. patent application Ser. No. 16/893,299, filed 4 Jun. 2020, titled “GRAPH EXPANSION AND OUTCOME DETERMINATION FOR GRAPH-DEFINED PROGRAM STATES.” This patent also claims the benefit of U.S. Provisional Patent Application 63/052,329, filed 15 Jul. 2020, titled “EVENT-BASED ENTITY SCORING IN DISTRIBUTED SYSTEMS.” This patent also claims the benefit of U.S. Provisional Patent Application 63/053,217, filed 17 Jul. 2020, titled “CONFIDENTIAL GOVERNANCE VERIFICATION FOR GRAPH-BASED SYSTEM.” This patent also claims the benefit of U.S. Provisional Patent Application 63/055,783, filed 23 Jul. 2020, titled “HYBRID DECENTRALIZED COMPUTING ENVIRONMENT FOR GRAPH-BASED EXECUTION ENVIRONMENT.” This patent also claims the benefit of U.S. Provisional Patent Application 63/056,984, filed 27 Jul. 2020, titled “MULTIGRAPH VERIFICATION.” The entire content of each aforementioned patent filing is hereby incorporated by reference.
BACKGROUND 1. FieldThis disclosure relates generally to computer systems and, more particularly, to graph-manipulation based domain-specific execution environments.
2. BackgroundDistributed applications operating on a distributed computing platform may be useful in a variety of contexts. Such applications can store program state data on a tamper-evident ledger operating on the distributed computing platform. The use of a tamper-evident ledger or some other data systems distributed over multiple computing devices may increase the security and reliability of distributed applications.
SUMMARYThe following is a non-exhaustive listing of some aspects of the present techniques. These and other aspects are described in the following disclosure.
Some aspects include a process that includes obtaining a directed graph encoding a symbolic artificial intelligence (AI) model used by a first entity, the directed graph comprising a first set of vertices and a first set of edges associating pairs of vertices of the first set of vertices. In some embodiments, each respective vertex of the first set of vertices is associated with a vertex status and is labeled with a category selected from a set of mutually-exclusive categories and a vertex of the first set of vertices is associated with a conditional statement that is indicated as triggerable by the first entity. The process may include determining a set of features based on the directed graph, the set of features comprising an identifier of a graph portion template, where each respective vertex of the graph portion template of the graph portion template is labeled with a same category from the set of mutually-exclusive categories as a corresponding respective vertex of a graph portion of the directed graph and is associated with a same count of edges. The process may include obtaining a set of model parameter values for a machine learning model based on the graph portion template and providing the set of model parameter values and the graph portion templates to the first entity. In some embodiments, the set of model parameter values are used by to determine an outcome score based on the directed graph.
Some aspects include a tangible, non-transitory, machine-readable medium storing instructions that when executed by a data processing apparatus cause the data processing apparatus to perform operations including the above-mentioned process.
Some aspects include a system, including: one or more processors; and memory storing instructions that when executed by the processors cause the processors to effectuate operations of the above-mentioned process.
The above-mentioned aspects and other aspects of the present techniques will be better understood when the present application is read in view of the following figures in which like numbers indicate similar or identical elements:
While the present techniques are susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. The drawings may not be to scale. It should be understood, however, that the drawings and detailed description thereto are not intended to limit the present techniques to the particular form disclosed, but to the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present techniques as defined by the appended claims.
DETAILED DESCRIPTION OF CERTAIN EMBODIMENTSTo mitigate the problems described herein, the inventors had to both invent solutions and, in some cases just as importantly, recognize problems overlooked (or not yet foreseen) by others in the field of program testing. Indeed, the inventors wish to emphasize the difficulty of recognizing those problems that are nascent and will become much more apparent in the future should trends in industry continue as the inventors expect. Further, because multiple problems are addressed, it should be understood that some embodiments are problem-specific, and not all embodiments address every problem with traditional systems described herein or provide every benefit described herein. That said, improvements that solve various permutations of these problems are described below.
Technology-based self-executing protocols, such as smart contracts and other programs, allow devices, sensors, and program code have seen increased use in recent years. However, some smart contracts and contract information models often rely on program instructions or industry-specific data structures, which may be difficult to generalize, use for comparison analysis, or reuse in similar contexts due to minor differences in contract details. As a result, uses of smart contracts has not extended into areas that are often the domain of natural language documents. Described herein is a process and related system to construct, interpret, enforce, analyze, and reuse terms for a smart contract in a systematic and unambiguous way across a broad range of applicable fields. In contrast, contracts encoded in natural language text often rely on social, financial, and judicial systems to provide the resources and mechanisms to construct, interpret, and enforce terms in the contracts. As contract terms increase in number or a situation within which the contract was formed evolves, such a reliance may lead to a lack of enforcement, ambiguity, and wasted resources spent on the re-interpretation or enforcement of contract terms.
Some embodiments may include smart contracts (or other programs) that include or are otherwise associated with a directed graph representing a state of the smart contract. In some embodiments, vertices of the graph may be associated with (e.g., encode, or otherwise represent) norms (e.g., as norm objects described below) of the smart contract, like formal language statements with a truth condition paired with a conditional statement (sometimes known as a “conditional”) that branches program flow (and changes norm state) responsive to the whether the truth condition is satisfied, for instance, “return a null response if and only if an API request includes a reserved character in a data field.” In some embodiments, norms of a smart contract may represent terms of a contract being represented by the smart contract, legal conditions of the contract, or other verifiable statements. As used herein, a smart contract may be a self-executing protocol executable as a script, an application, or portion of an application on a distributed computing platform, centralized computing system, or single computing device. Furthermore, as used herein, a graph may be referred to as a same graph after the graph is manipulated. For example, if a graph being referred to as a “first graph” is represented by the serialized array [[1,2], [2,3], [3,4]] is modified to include the extra vertex and graph edge “[1,5]” and become the modified graph represented by the serialized array “[[1,2], [2,3], [3,4], [1,5]],” the term “first graph” may be used to refer to the modified graph. Additionally, it should be understood that a data structure need not be labeled in program code as a graph to constitute a graph for the present purposes, as long as that data structure encodes the relationships between values described herein. For example, a graph may be encoded in a key-value store even if source code does not label the key-value store as a graph.
A self-executing protocol may be a program, like a smart contract. Self-executing protocols may execute responsive to external events, which may include outputs of third-party programs, and human input via a user interface. A self-executing protocol may execute on a computing substrate that involves human intervention to operate, like turning on a computer and launching an event listener.
A norm of a smart contract may be encoded in various formal languages (like programming languages, such as data structures encoding statements in a domain-specific programming language) and may include or otherwise be associated with one or more conditional statements, a set of norm conditions of the one or more conditional statements, a set of outcome subroutines of the one or more conditional statements, a norm status, and a set of consequent norms. In some embodiments, satisfying a norm condition may change a norm status and lead to the creation or activation of the consequent norms based on the actions performed by the system when executing the outcome subroutines corresponding to the satisfied norm condition. In some embodiments, a norm may be triggered (i.e. “activated”) when an associated norm condition is satisfied by an event, as further described below. Alternatively, some types of norms may be triggered when a norm condition is not satisfied before a condition expiration threshold is satisfied. As used herein, a triggerable norm (i.e. “active norm”) is a norm having associated norm conditions may be satisfied by an event. In contrast, a norm that is set as not triggerable (i.e. “inactive”) is a norm is not updated even if its corresponding norm conditions are satisfied. As used herein, deactivating a norm may include setting the norm to not be triggerable.
A smart contract and its norms may incorporate elements of a deontic logic model. A deontic logic model may include a categorization of each of the norms into one of a set of deontic primitive logical categories. A deontic primitive logical category (“logical category”) may include a label such as “right,” “obligation,” or “prohibition.” The logical category may indicate a behavior of the norm when the norm is triggered. In addition, a norm of the smart contract may have an associated norm status such as “true,” “false,” or “unrealized,” where an event may trigger a triggerable norm by satisfying a norm condition (and thus “realizing” the norm). These events may be collected into a knowledge list. The knowledge list may include an associative array of norms, their associated states, an initial norm status during the initial instantiation of the associated smart contract, their norm observation times (e.g., when a norm status was changed, when an event message was received, or the like), or other information associated with the norms. The smart contract may also include a set of consequent actions, where a consequent action may include an association between a triggered norm and any respective consequent norms of the smart contract. As further discussed below, the set of consequent actions may be updated as events occur and the smart contract state is updated, which may result in the formation of a history of previous consequent actions. It should be understood that the term “norm” is used for illustrative purposes and that this term may have different names in other references and contexts. The labeling of norms may also be used for symbolic artificial intelligence (AI) systems. As described further below, the use of these symbolic AI systems in the context of a smart contract may allow for sophisticated verification and predictive techniques that may be impractical for pure neural network systems which do not use symbolic AI systems. It should be understood that, while the term “logical category” is used in some embodiments, other the terms may be used for categories or types of categories without loss of generality. For example, some embodiments may refer to the use of a “category label” instead of a logical category.
Some embodiments may store a portion of the smart contract state in a data serialization format (“serialized smart contract state data”). For example, as further described below, some embodiments may store a vertices of a directed graph (or both vertices and edges) in a data serialization format. In response to determining that an event has occurred, some embodiments may deserialize the serialized smart contract state data into a deserialized directed graph. In some embodiments, a vertex (a term used interchangeably with the term node) of the directed graph may be associated with a norm from a set of norms of the smart contract and is described herein as a “norm vertex,” among other terms, where a norm vertex may be connected one or more other norm vertices via graph edges of the directed graph. Some embodiments may then update the directed graph based on a set of consequent norms and their associated consequent norm vertices, where each of the consequent norms are determined based on which norms were triggered by the event and what norm conditions are associated with those active norms. The updated directed graph may then be reserialized to update the smart contract. In some embodiments, a norm vertex may not have any associated conditions. In some embodiments, the amount of memory used to store the serialized smart contract state data may be significantly less than the memory used by deserialized smart contract state data. During or after the operation to update the smart contract, some embodiments may send a message to entities listed in a list of entities (such as an associative array of entities) to inform the entities that the smart contract has been updated, where the smart contract includes or is otherwise associated with the list of entities. Furthermore, it should be understood in this disclosure that a vertex may include (or comprise) a condition by being associated with the condition. For example, a norm vertex may include a first norm condition by including a reference pointer to the first norm condition.
In some embodiments, generating the smart contract may include using an integrated development environment (IDE) and may include importing libraries of provisions re-used across agreements. Furthermore, some embodiments may generate a smart contract based on the use of natural language processing (NLP), as further described below. For example, some embodiments may apply NLP operations to convert an existing prose document into a smart contract using operations similar to those described for patent application 63/034,255, titled “Semantic Contract Maps,” which is herein incorporated by reference. For example, some embodiments may apply a set of linear combinations of feature observations and cross observations across first order and second orders in feature space to determine a smart contract program or other symbolic AI program. Alternatively, or in addition, some embodiments may include constructing a smart contract from a user interface or text editor without using an existing prose document. In some embodiments, the smart contract may be encoded in various forms, such as source code, bytecode, or machine code encodings. In some embodiments, a smart code may be generated or modified in one type of encoding and be converted to another type of encoding before the smart code is used. For example, a smart contract may be edited in a source code encoding, and the smart contract may be executed by converting the smart contract into a bytecode encoding executing on a distributed computing platform. As used herein, a smart contract may be referred to as a same smart contract between different encodings of the smart contract. For example, a smart contract may be written in source code and then converted to a machine code encoding, may be referred to as a same smart contract.
Furthermore, as used herein, the sets of items of a smart contract data model may be encoded in various formats. A set of items be encoded in an associative array, a b-tree, a R-tree, a stack, or various other types of data structures. As used herein, the sets of items in the data model may be determined based on their relationships with each other. For example, a set of entities may be encoded as an associative array of entities or may be encoded as an entities b-tree, and elements of a knowledge list may include references to an entity in the set of entities for either type of encoding. In some embodiments, sets of items in their respective data models may be based on the underlying relationships and references between the items in the sets of items, and embodiments should not be construed as limited to specific encoding formats. For example, while some embodiments may refer to an associative array of norms, it should be understood that other embodiments may use a b-tree to represent some or all of the set of norms.
A smart contract may be stored on different levels of a memory hierarchy. A memory hierarchy of may include (in order of fastest to slowest with respect to memory access speed) processor registers, Level 0 micro operations cache, Level 1 instructions cache, Level 2 shared cache, Level 3 shared cache, Level 4 shared cache, random access memory (RAM), a persistent flash memory, hard drives, and magnetic tapes. For example, a Level 1 cache of a computing device may be faster than a RAM of the computing device, which in turn may be faster than a persistent flash memory of the computing device. In some embodiments, the memory of a computing device at a first layer of the memory hierarchy may have a lower memory capacity than a memory of the computing device at a slower layer of the memory hierarchy. For example, a Level 0 cache may memory capacity of 6 kibibytes (KiB), whereas a Level 4 cache may have a memory capacity of 128 mebibytes (MiB). In some embodiments, memory may be further distinguished between persistent storage and non-persistent storage (i.e. “non-persistent memory”), where persistent storage is computer memory that may retain the values stored in it without an active power source. For example, persistent storage may include persistent flash memory, hard drives, or magnetic tape, and non-persistent memory may include processor registers, cache memory, or dynamic RAM. In some embodiments, a smart contract may be stored on memory at different levels of the memory hierarchy to increase storage efficiency of the smart contract. For example, serialized smart contract state data of the smart contract may be stored on RAM of a computing device while the deserialized smart contract state data may be stored on a cache of the computing device.
In some embodiments, the smart contract may update infrequently, such as less than once per hour, less than once day, less than once per month, or the like. The relative infrequency of the updates can mean that the relative computing resources required to deserialize and reserialize data be significantly less than the computing resources required to maintain deserialized data in higher-speed memory. In some embodiments, the dynamic program state By serializing a portion of the smart contract data and persisting the serialized data instead of the corresponding deserialized data to a persistent storage, a computing system may use reduce the memory requirements of storing and executing the smart contract. In addition, the computing system may also increase the number of smart contracts being executed concurrently by a distributed computing platform or single computing device. Furthermore, as used herein, updating a value may include changing the value or generating the value.
As described herein, some embodiments may store smart contract data in other forms. For example, while some embodiments may temporarily store a directed graph in non-persistent storage, some embodiments may store the directed graph on a persistent storage. In some embodiments, various other types of information such as norm statuses (e.g. “triggered,” “failed,” “satisfied,” etc.) or logical categories (e.g. “rights,” “obligation,” “prohibition,” etc.) may be included in or otherwise associated with some or all of the vertices of the directed graph. Furthermore, some embodiments may generate visual display representing of the program state data to show the directed graph and its associated statuses, categories, or other information. For example, as further described below, some embodiments may display the directed graph as a hierarchical visual element such as a hierarchy tree in a web application.
A smart contract may be implemented in various ways. For example, some embodiments may construct, enforce, or terminate the smart contract using a distributed ledger or distributed computing system. Alternatively, some embodiments may implement the smart contract using a request-response system over a public or private internet protocol (IP) network. Use of the methods described herein may increase the efficiency of smart contract enforcement by advancing the state of complex multi-entity agreements in a fast and unambiguous way. Furthermore, implementing and using smart contracts with the embodiments described herein may allow for the comparison, quantification, and reuse of smart contracts in a way that would be inapplicable to custom-coded smart contracts.
In some embodiments, the smart contract may be stored in a tamper-evident data-store. As discussed below, tamper-evident data stores (e.g., repositories rendering data tamper-evident with one or more tamper-evident data structures) afford desirable properties, including making it relatively easy to detect tampering with entries in the data store and making it relatively difficult or impossible to tailor entries to avoid such detection. Furthermore, various smart contracts may be operating across one or more nodes of the tamper-evident data store, reducing the susceptibility of the smart contract to regional disturbances.
None of the preceding should be taken to suggest that any technique is disclaimed or that the approaches described herein may not be used in conjunction with other approaches having these or other described disadvantages, for instance, some embodiments may use a custom-written smart-contract that includes one or more of the norms, data structures, or graphs described herein. Or some embodiments may store a directed graph without serialization or deserialization operations. Or some embodiments may be implemented on a centralized server without storing smart contract state data on a distributed computing system such as a decentralized computing system. Further, it should be emphasized that the data structures, concepts, and instructions described herein may bear labels different from those applied here in program code, e.g., a data structure need not be labeled as a “node” or a “graph” in program code to qualify as such, provided that the essential characteristics of such items are embodied.
In some embodiments, the processes and functionality described herein may be implemented as computer code stored on a tangible, non-transitory, machine-readable medium, such that when instructions of the code are executed by one or more processors, the described functionality may be effectuated. For example, the process 100 of
In some embodiments, the process 100 includes determining that an event has occurred based on an event message, a calculation, or a condition expiration threshold, as indicated by block 104. In some embodiments, the system may determine that an event has occurred after receiving an event message at an API of the system indicating that the event has occurred. As used herein, an event message may be transmitted across or more packets over a wired or wireless connection, where a system may continuously, periodically, or be activated to listen for an event message. In some embodiments, as described further below, an event message may be transmitted over a public or private IP network. Alternatively, or in addition, the event message may be transmitted via the channels of a distributed computing system. For example, the event message may be transmitted from a first node of a distributed computing system (e.g., a blockchain platform) to a second node of the distributed computing system, where the first node and second node may be at different geographic locations (e.g., different nodes executing on different computing devices) or share a same geographic location (e.g., different nodes executing on a same computing device). Furthermore, an event message may be sent by a first smart contract executing on a first computing distributed platform to a second smart contract executing on a same or different distributed computing platform. In some embodiments, determining than event has occurred does not require verification that the event has occurred. For example, in some embodiments, receiving an event message indicating an event has occurred may be sufficient for the system to determine that the event occurred. Furthermore, in some embodiments, a norm vertex may be triggered based on an event satisfying a subset of its associated norm conditions. Alternatively, a norm vertex may be triggered only after an event satisfies all of its associated norm conditions.
In some embodiments, the event may include satisfying a condition expiration threshold associated with a triggerable norm vertex (herein “triggerable vertex”) without satisfying a norm condition associated with the triggerable vertex, where a norm condition may be various types of conditions implemented in a computer-readable form to return a value (e.g., “True,” “False,” set of multiple binary values, or the like). For example, a norm condition may include an “if” statement to test whether a payload containing a set of values was delivered to an API of the system by a specific date, where a condition expiration threshold is associated with the norm condition. After the specific date is reached, the system may determine that the condition expiration threshold is satisfied and determine whether the associated norm condition is satisfied. In response to a determination that the norm condition is not satisfied, the system may determine that an event has occurred, where the event indicates that a condition expiration threshold associated with a triggered norm vertex (herein “triggered vertex”) is satisfied and that an associated norm condition of the triggered vertex is not satisfied. As further stated below, such an event may trigger the associated norm vertex and result in the activation of a set of norms, where the activation of the set of norms may be represented by the generation or association of an adjacent vertex to the triggered vertex, where the adjacent vertex may be updated to be triggerable. As used in this disclosure, it should be understood that satisfying the condition expiration threshold of a triggerable vertex does satisfy a condition associated with the triggerable vertex.
In some embodiments, the event message may include a publisher identifier to characterize a publisher of the event message. As used herein, a publisher may be an entity and may include various sources of an event message. For example, a publisher may include a publisher in a publisher-subscriber messaging model or a sender of a response or request in a response-request messaging model. In some embodiments, the publisher identifier may be an entity identifier that is a specific name unique to a source of the event message. For example, a publisher identified by the publisher identifier “BLMBRG” may be transmitted in the event message, where “BLMBRG” is unique to a single publisher. Alternatively, or in addition, a publisher identifier may include or be otherwise associated with an identifier corresponding to an entity type that may be assigned to one or more sources of event messages. For example, the publisher identifier may include or otherwise be associated with an entity type such as “TRUSTED-VENDOR,” “ADMIN”, or the like.
After receiving a publisher identifier, the system may determine whether the publisher identifier is associated with one of a set of authorized publishers with respect to the event indicated by the event message. In some embodiments, the system may refer to a set of authorized publishers corresponding to the event indicated by the event message. For example, the event message may indicate that an event associated with the event message “PAY DELIVERED” has occurred. In in response, the system may determine that the event satisfies an condition threshold, where satisfying the condition threshold may include a determination that the event satisfies one or more norm conditions in an associative array of conditions and that the associated publisher is authorized to deliver the message. The associative array of conditions may include a list of norm conditions that, if satisfied, may result in triggering at least one triggerable vertex of the smart contract. For example, the system may determine that the event “PAY DELIVERED” is a direct match with the norm condition “if(PAY DELIVERED)” of the associative array of conditions. In some embodiments, the system may then refer to the set of authorized publishers associated with the event “PAY DELIVERED.” The system may then determine whether the publisher identifier is in the set of authorized publishers or otherwise associated with the set of authorized publishers, such as by having an entity type representing the set of authorized publishers. In some embodiments, if the system determines that the event message is not authorized, the event message may be rejected as not authorized.
In some embodiments, the operation to authorize the event may include a operations represented by Statement 1 or Statement 2 below, where “prop” may be a string value including an event and “pub” may be a string value representing a publisher identifier or entity type. In some embodiments, Statement 1 below may represent an authorization operation that includes the arrival of an event E[pub] from publisher pub. The system may then compare the publisher “P[E[pub]]” of the event “E[pub]” with each of a set of authorized publishers “D[E[prop]][pub]”, where each of the set of authorized publishers is authorized to publish the event “E[prop]”. In some embodiments, the set of entities may include or otherwise be associated with the set of authorized publishers. Statement 2 may represent the situation which a plurality of entities may publish a valid event and the systems authorizes a message based on the entity type “P[E[pub]][role]” being in the set of authorized publishers “D[E[prop]][pub],” where the set of authorized publishers “D[E[prop]][pub]” may include authorized publisher type:
D[E[prop]][pub]==P[E[pub]] (1)
D[E[prop]][pub]==P[E[pub]][role] (2)
In some embodiments, the set of authorized publishers may include a set of publisher identifiers, and the publisher identifier may in the set of publisher identifiers. For example, if the publisher identifier is “BLMBRG” and the set of authorized publishers include “BLMBRG,” the system may determine that an event message including the publisher identifier “BLMBRG” is authorized. Alternatively, or in addition, the set of authorized publishers may include one or more authorized entity types and a respective publisher may be an authorized publisher if the respective publisher identifier is associated with the authorized entity type. For example, if the publisher identifier is “BLMBRG,” and if the set of authorized publishers include the entity type “AUTH_PROVIDERS,” and if “BLMBRG” is associated with “AUTH_PROVIDERS” via an associative array, then the system may determine that the publisher identifier is associated with the set of authorized publishers. In response, the system may determine that the event message including the publisher identifier “BLMBRG” is authorized. In some embodiments, the system may determine that one or more events indicated by the event message has occurred only after determining that the event message is authorized.
In some embodiments, the event message may include a signature value usable by the system to compute a cryptographic hash value. Furthermore, some event messages may include the event payload with the signature value (e.g., via string concatenation) to compute the cryptographic hash value. The system may use various cryptographic hashing algorithms such as SHA-2, Bcrypt, Scrypt, or the like may be used to generate a cryptographic hash value. In some embodiments, the system may use salting operations or peppering operations to increase protection for publisher information. In some embodiments, the system may retrieve a cryptographic certificate based on a publisher identifier as described above and authenticate the event message after determining that on the cryptographic hash value satisfies one or more criteria based on the cryptographic certificate. A cryptographic certificate may include a cryptographic public key used to compare with the cryptographic hash value, as further discussed below. In addition, the cryptographic certificate may also include one or more second cryptographic values indicating a certificate issuer, certificate authority private key, other certificate metadata, or the like.
In some embodiments, a smart contract may include or be associated with a plurality of cryptographic certificates. The system may determine of which cryptographic certificate to use may be based on a map of entities of the smart contract. In some embodiments, the operation to authenticate the event may include a statement represented by Statement 3 below, where “v” may represent a signature verification algorithm, E[sig] may represent a signature value of an event object “E,” “P[E[pub]]” may represent a data structure that includes the entity that had published the event E, and P[E[pub]][cert] may represent a cryptographic certificate value such as cryptographic public key:
v(E[sig],P[E[pub]][cert])==True (3)
Various signature verification algorithms may be used to authenticate an event message based on a signature value of the event message. For example, the system may determine that the cryptographic hash value is equal to the cryptographic certificate, and, in response, authenticate the event message. In some embodiments, the system may determine that one or more events indicated by the event message has occurred only after authenticating the event message.
In some embodiments, the system may determine that an event has occurred based on a determination that a condition expiration threshold has been reached. One or more norms represented by norm vertices in the smart contract may include a condition expiration threshold such as an obligation that must be fulfilled by a first date or a right that expires after a second date. For example, a smart contract instance executing on the system may include a set of condition expiration thresholds, where the set of condition expiration thresholds may include specific dates, specific datetimes, durations from a starting point, other measurements of time, other measurements of time intervals, or the like. The system may check the set of condition expiration thresholds to determine if any of the condition expiration thresholds have been satisfied.
An event message may be transmitted under one of various types of messaging architecture. In some embodiments, the architecture may be based on a representational state transfer (REST) system, where the event message may be a request or response. For example, a system may receive a request that includes the event message, where the request includes a method identifier indicating that the event message is stored in the request. As an example, the system may receive a request that includes a “POST” method indicator, which indicates that data is in the request message. In addition, the request my include a host identifier, where the host identifier indicates a host of the smart contract being executed by the system. For example, the host identifier may indicate a specific computing device, a web address, an IP address, a virtual server executing on a distribute computing platform, a specific node of a decentralized computing system, or the like.
In some embodiments, the architecture may be based on a publisher-subscriber architecture such as the architecture of the advanced message queuing protocol (AMQP), where the event message may be a either a publisher message or subscriber message. For example, using the AMQP, a client publisher application may send an event message over a TCP layer to an AMQP server. The event message may include a routing key, and the AMQP server may act as a protocol broker that distributes the event message to the system based on the routing key after storing the event message in a queue. In some embodiments, the system may be a subscriber to the client publisher application that sent the event message.
In some embodiments, the process 100 includes determining which smart contracts of a set of active smart contracts that will change state based on the event, as indicated by block 108. As discussed above, in some embodiments, the system may determine that the event satisfies one or more norm conditions, and, in response, determine that the instance of the smart contract will change state. For example, as further discussed below, the system may determine that the event indicated “PAYLOAD 0105 PROVIDED” satisfies the norm condition represented by the condition “IF DELI VERED(PAYLOAD),” In response, the system may determine that the smart contract will change state. Alternatively, or in addition, as discussed above, the system may determine that the event does not satisfy one or more norm conditions but does satisfy a condition expiration threshold. In response, the system may determine that the instance of the smart contract will change state based on the event not satisfying one or more norm conditions while having satisfied the condition expiration threshold. Furthermore, while this disclosure may recite the specific use of a smart contract program in certain sections, some embodiments may use, modify, or generate other symbolic AI programs in place of a smart contract, where symbolic AI programs are further discussed below.
In some embodiments, the system may include or otherwise have access to a plurality of smart contracts or smart contract instances. The system may perform a lookup operation to select which of the smart contracts to access in response to determining that an event has occurred. In some operations, the smart contract may compare an event to the associative array of conditions corresponding to each of a set of smart contracts to select of the set of smart contracts should be updated and filter out smart contracts that would not change state based on the event. The system may then update each of the smart contract instances associated with a changed norm status, as discussed further below. Furthermore, the system may then update the respective associative array of conditions corresponding to the set of smart contracts. In some embodiments, an associative array of conditions may include only a subset of norm conditions associated with a smart contract, where each the subset of norm conditions is associated with a triggerable vertex of the smart contract. In some embodiments, the system may first deduplicate the norm conditions before performing a lookup operation to increase performance efficiency. For example, after determining that an event has occurred, some embodiments may search through a deduplicated array of norm conditions. For each norm condition that the event would trigger, the system may then update the one or more smart contracts associated with the norm condition in the deduplicated array of norm conditions. By selecting smart contracts from a plurality of smart contracts based on an array of norm conditions instead of applying the event to the norm conditions associated with the norm vertices of each of the set of smart contracts, the system may reduce computations required to update a set of smart contracts.
The smart contract or associated smart contract state data may be stored on various types of computing systems. In some embodiments, the smart contract state data may be stored in a centralized computing system and the associated smart contract may be executed by the centralized computing system. Alternatively, or in addition, the smart contract or associated smart contract state data may be stored on a distributed computing system (like a decentralized computing system) and the associated smart contract may be executed using a decentralized application. Fore example, the smart contract may be stored on and executed by a Turing-complete decentralized computing system operating on a set of peer nodes, as further described below.
In some embodiments, the smart contract data may include or be otherwise associated with a set of entities, such as a set of entities encoded as an associative array of entities. The associative array of entities that may include one or more entities that may interact with or view at least a portion of the data associated with the smart contract. In some embodiments, the associative array of entities may include a first associative array, where keys of the first associative array may indicate specific smart contract entities (e.g. data observers, publishers, or the like), and where each of the keys may correspond with a submap containing entity data such as a full legal name, a legal identifier such as a ISIN/CUSIP and an entity type of the entity such as “LENDER,” “BORROWER”, “AGENT,” “REGULATOR,” or the like. In some embodiments, one or more entities of the associative array of entities may include or be associated with a cryptographic certificate such as a cryptographic public key. As described above, the cryptographic certificate may be used to authenticate an event message or other message. By including authorization or authentication operations, the system may reduce the risk that an unauthorized publisher sends an event message or that the event message from a publisher is tampered without the system determining that tampering had occurred. In addition, authorization or authentication operations increase the non-repudiation of event messages, reducing the risk that a publisher may later disclaim responsibility for transmitting an event message.
In some embodiments, the smart contract may also include or otherwise be associated a set of conditions, such as a set of conditions encoded as an associative array of conditions. In some embodiments, the associative array of conditions may include a set of norm conditions and associated norm information. In some embodiments, the set of norm conditions may be represented by an associative array, where a respective key of the associative array may be a respective norm condition or norm condition identifier. The corresponding values of the associative array may include a natural language description of the corresponding condition and one or more publisher identifiers allowed to indicate that an event satisfying the respective norm condition has occurred. In some embodiments, the publisher identifier may indicate a specific entity key or an entity type. Furthermore, the smart contract may also include or otherwise be associated with a set of norm vertices or a set of graph edges connecting the vertices, as further described below.
In some embodiments, the process 100 includes deserializing a serialized array of norm vertices to generate a deserialized directed graph, as indicated by block 112. In some embodiments, the smart contract may include or otherwise be associated with a set of norm vertices encoded as a serialized graph in various data serialization formats, where the smart contract may encode a part or all the norm vertices by encoding the graph edges connecting the norm vertices The serialized graph may include a representation of an array of subarrays. A data serialization format may include non-hierarchical formats or flat-file formats, and may be stored in a persistent storage. In some embodiments, a serialized array of norm vertices may include numeral values, strings, strings of bytes, or the like. For example, the array of norm vertices (or other data structures in program state) may be stored in a data serialization format such as JSON, XML, YAML, XDR, property list format, HDF, netCDF, or the like. For example, an array may be decomposed into lists or dictionaries in JSON amenable to serialization. Each subarray of an array of subarrays may include a pair of norm vertices representing a directed graph edge. For example, a subarray may include a first value and a second value, where the first value may represent a tail vertex of a directed graph edge, and where the second value may represent a head vertex of the directed graph edge. For example, a subarray may include the value “[1,5]” where the first value “1” represents a tail vertex indicated by the index value “1” and “5” represents a head vertex indicated by the index value “5.” While in serialized form, the array of norm vertices may reduce memory requirements during data storage operations and bandwidth requirements during data transfer operations.
In some embodiments, the serialized array of norm vertices may be used to construct an adjacency matrix or an index-free adjacency list to represent a deserialized directed graph during a deserialization operation. In some embodiments, an adjacency matrix or adjacency list may increase efficient graph rendering or computation operations. In some embodiments, the deserialized directed graph may be stored in a faster layer of memory relative to the serialized graph, such as in a non-persistent memory layer. For example, the system may deserialize a serialized array of vertices stored in flash memory to a deserialized directed graph stored in Level 3 cache. In some embodiments, as further described below, instead of forming a directed graph that includes all of the norm vertices included in the serialized array of norm vertices, the system may instead form a directed graph from a subset of the serialized array of norm vertices. As described above, each norm vertex may have an associated norm status indicating whether the norm vertex is triggerable. In response, the system may form a directed graph of the triggerable vertices without rendering or otherwise processing one or more norm vertices not indicated to be triggerable. Using this method, a vertex that is included in the serialized array of vertices may be absent in the directed graph stored in non-persistent memory. By reducing the number of number of vertices in a deserialized directed graph, the efficiency of querying and updating operations of the smart contract may be increased.
In some embodiments, the system may include an initial set of norm vertices that is distinct from the array of norm vertices. For example, some embodiments may determine that the smart contract had made a first determination that an event had occurred. In some embodiments, the system may search the data associated with the smart contract to find an initial set of norm vertices representing an initial state of the smart contract. The system may then deserialize the initial set of norm vertices when executing the smart contract and perform the operations further described below. The system may then deserialize a different array of norm vertices during subsequent deserialization operations.
In some embodiments, the process 100 includes determining a set of triggerable vertices based on the directed graph, as indicated by block 120. In some embodiments, the system may determine the set of triggerable vertices based on the directed graph stored in non-persistent memory by searching through the vertices of the directed graph for each of the head vertices of the directed graph and assigning these vertices as a set of head vertices. The system may then search through the set of head vertices and filter out all head vertices that are also tail vertices of the directed graph, where the remaining vertices may be the set of leaf vertices of the directed graph, where each of the leaf vertices represent a triggerable vertex. Thus, the set of leaf vertices determined may be used as the set of triggerable vertices.
Alternatively, in some embodiments, a vertex of the set of norm vertices may include or otherwise be associated with a norm status indicating whether the vertex is triggerable or not. In some embodiments, the system may search through the directed graph for vertices that have an associated norm status indicating that the respective vertex is triggerable. Alternatively, or in addition, the system may search through a list of norm statuses associated with the vertices of the serialized array of norm vertices to determine which of the vertices is triggerable and determine the set of triggerable vertices. For example, in some embodiments, each norm vertex of a smart contract may have an associated norm status indicating whether the vertex is triggerable or not triggerable, where the vertices and their associated statuses may be collected into a map of vertex trigger states. The system may then perform operations to traverse the map of vertex trigger states and determine the set of triggerable vertices by collecting the vertices associated with a norm status indicating that the vertex is triggerable (e.g. with a boolean value, a numeric value, a string, or the like). For example, the system may perform operations represented by Statement 4 below, where G may represent a graph and may be an array of subarrays g, where each subarray g may represent a norm vertex and may include a set of values that include the value assigned to the subarray element g[4], where the subarray element g[4] indicates a norm status, and “Active” indicates that the norm vertex associated with subarray g is triggerable, and A is the set of triggerable vertices:
A←{g∈G|g[4]=“Active”} (4)
In some embodiments, the process 100 includes determining a set of triggered vertices based on the set of triggerable vertices, as indicated by block 124. In some embodiments, the system may compare determine the set of triggered vertices based on which the norm conditions associated with the vertices of the directed graph are satisfied by the event. In some embodiments, a norm condition may directly include satisfying event. For example, a norm condition may include “IF DELIVERED(PAYMENT),” where the function “DELIVERED” returns a boolean value indicating whether a payment represented by the variable “PAYMENT” is true or false. The system may then determine that the norm condition is satisfied if “DELIVERED(PAYMENT)” returns the boolean value “True.” The system may then add the vertex associated with the norm condition to the set of triggered vertices. For example, the system may perform operations represented by Statement 5 below, where “A” is the set of triggerable vertices determined above, and where each subarray “a” may represent a triggerable vertex and may include a set of values that include the value assigned to the subarray element a[1], where the subarray element a[1] indicates a condition, and “U” is the set of triggered vertices, and “N” is an associative array that describes the possible graph nodes that may be triggered, such that, for an event prop, N[prop] may return a structure that contains defining details of the vertices associated with the event prop:
U←{a∈A|N[a[1]][prop]=E[prop]} (5)
In some embodiments, the determination that an event satisfies a norm condition may be based on a categorization of a norm into logical categories. As further described below in
In some embodiments, a snapshot contract status may be associated with the smart contract and may be used to indicate a general state of the smart contract. The snapshot contract status may indicate whether the obligations of a contract are being fulfilled or if any prohibitions of the contract are being violated. For example, in some embodiments, satisfying an obligation norm condition may result in an increase in the snapshot contract status and triggering a prohibitions norm may result in a negative change to the snapshot contract status.
In some embodiments, the process 100 includes performing one or more operations indicated by blocks 152, 154, 156, and 160 for each of the respective triggered vertex of the set of triggered vertices, as indicated by block 150. In some embodiments, the process 100 includes updating the respective triggered vertex based on an event by updating a norm status associated with the respective triggered vertex, as indicated by block 152. Updating a respective triggered vertex may include updating one or more norm statuses or other status values associated with the respective triggered vertex. For example, a norm status of the respective triggered vertex may be updated to include one of the strings “SATISFIED,” “EXERCISED,” “FAILED,” or “CANCELED,” based on the norm conditions associated with the respective triggered vertex having been satisfied, exercised, failed, or canceled, respectively. In some embodiments, the system may update a norm status to indicate that the respective triggered vertex is not triggerable. For example, an obligation norm of a smart contract may be required to be satisfied only once. In response, after determining that the norm condition associated with the obligation has been satisfied by an event, the system may update a first status value associated with the respective triggered vertex to “false,” where the first status value indicates whether the respective triggered vertex is triggerable. In some embodiments, the one or more status values may include a valence value indicating the number of connections from the respective triggered vertex to another vertices, the number of connections to the respective triggered vertex from other vertices, or the like. As further described below, in some embodiments, the valence value or other status value associated with the respective triggered vertex may be updated after performing operations associated with the adjacent vertices of the respective triggered vertex.
In some embodiments, the process 100 includes determining whether a respective adjacent vertex of the respective triggered vertex should be set to be triggerable, as indicated by block 154. In some embodiments, the respective triggered vertex may include a pointer to or otherwise be associated with a set of adjacent vertices, where each of the set of adjacent vertices represent a norm of the smart contract that are set to occur after the respective triggered vertex is triggered. In some embodiments, the system may determine whether an adjacent vertex of a respective triggered vertex should be set as triggerable based on specific conditions associated with the adjacent vertex. For example, a respective triggered vertex may include program code instructing that a first set of adjacent vertices should be set to be triggerable if a first set of conditions are satisfied and that a second set of adjacent vertices should be set to be triggerable if a second set of conditions are satisfied, where the first set of adjacent vertices are distinct from the second set of adjacent vertices. Alternatively, or in addition, the respective triggered vertex may include program instructing that a third set of adjacent vertices should be set to be triggerable if the first set of conditions are not satisfied but an associated condition expiration threshold is satisfied.
In some embodiments, the process 100 includes updating the respective adjacent vertex based on the event, as indicated by block 156. Updating the respective adjacent vertex based on the event may include setting one or more norm statuses associated with the adjacent vertex to indicate that the respective adjacent vertex is triggerable. For example, after a determination that a respective adjacent vertex associated with a permission norm is to be set to be triggerable, a norm status associated with the respective adjacent vertex may be updated to the value “triggerable.”
In some embodiments, the process 100 includes determining whether any additional triggered vertices are available, as indicated by block 160. In some embodiments, the system may determine that additional triggered vertices are available based on a determination that an iterative loop used to cycle through each the triggered vertices has not reached a termination condition. In response to a determination that additional triggered vertices are available, the process 100 may return to the operations of block 150. Otherwise, operations of the process 100 may proceed to block 164.
In some embodiments, the process 100 includes updating the directed graph based on the updated triggered vertices or the respective adjacent vertices, as indicated by block 170. In some embodiments, updating the directed graph may include updating an adjacency matrix or adjacency list representing the directed based on each of the triggered vertices or their respective adjacent vertices. In some embodiments, instead of looping through each updated vertex and then updating the directed graph, the system may update the directed graph during or after each update cycle. For example, after updating the respective triggered vertex as described in block 156, the system may update the deserialized directed graph.
In some embodiments, the process 100 includes updating the serialized array of norm vertices or other smart contract state data based on the directed graph and updated vertices, as indicated by block 174. In some embodiments, updating the serialized array of norm vertices may include serializing the directed graph into a data serialization format, as described above. In some embodiments, the data serialization format may be the same as the data serialization format used when performing operations described for block 112. For example, the system may implement a depth-first search (DFS) over the deserialized directed graph to record distinct edge pairs and update the serialized array of norm vertices by either modifying or replacing the serialized array of norm vertices.
In some embodiments, the system may update a knowledge set based on the event and smart contract state changes that occurred in response to the event. In some embodiments, the knowledge set may include a set of previous events. The set of previous events may be encoded as a list of previous events. The list of previous events may include a subarray, where each subarray includes an event identifier of a recorded event or information associated with the recorded event. For example, the list of previous events may include a date and time during which an event occurred, an event identifier, one or more norm conditions satisfied by the event, or the like. In some embodiments, a norm condition may be based on the list of previous events. For example, a norm condition may include a determination of whether an event type had occurred twice within a time duration based on the list of previous events. In some embodiments, the knowledge set may include a set of previously-triggered vertices, where the set of previously-triggered vertices may be encoded as an array of previously-triggered vertices. In some embodiments, the system may further update the knowledge set by updating the array of previously-triggered vertices based on the triggered vertices described above. For example, after updating a respective triggered vertex as described above, the system may update the array of previously-triggered vertices to include the respective triggered vertex. The array of previously-triggered vertices may include a vertex identifier associated with the respective triggered vertex, an event identifier associated with the event that triggered the respective triggered vertex, and a set of values identifying the vertices that are set to be triggerable after triggering the respective triggered vertex.
In some embodiments, the process 100 includes persisting the updated serialized array of norm vertices or other smart contract data to storage, as indicated by block 178. In some embodiments, persisting the smart contract data to storage may include updating the memory storage in a single computing device or a computing device of a centralized computing system. Alternatively, or in addition, persisting the smart contract data to storage may include storing the smart contract data to a decentralized tamper-evident data store. In some embodiments, by storing the serialized array of norm vertices in a decentralized tamper-evident data store instead of storing a deserialized directed graph in the decentralized tamper-evident data store, the system may increase the efficiency and performance of the data distribution amongst the nodes of the decentralized tamper-evident data store. Furthermore, in some embodiments, triggering a norm vertex may include triggering a smart contract termination action. When a smart contract termination action is triggered, vertices other than the respective triggered vertex may be updated to set the statuses of each vertex of these other vertices as not triggerable, even if these other vertices are not directly connected to the triggered vertex.
In some embodiments, the system may display a visualization of the smart contract state. For example, the system may display a visualization of smart contract state as a directed graph, such as (though not limited to) those shown in
In some embodiments, the process 100 or other processes described in this disclosure may execute on a decentralized computing platform capable of persisting state to a decentralized tamper-evident data store. Furthermore, in some embodiments, the decentralized computing platform may be capable of executing various programs, such as smart contracts, on the computing platform in a decentralized, verifiable manner. For example, each of a set of peer nodes of the computing platform may perform the same computations, and a consensus may be reached regarding results of the computation. In some embodiments, various consensus algorithms (e.g., Raft, Paxos, Helix, Hotstuff, Practical Byzantine Fault Tolerance, Honey Badger Byzantine Fault Tolerance, or the like) may be implemented to determine states or computation results of the various programs executed on the decentralized computing platform without requiring that any one computing device be a trusted device (e.g., require an assumption that the computing device's computation results are correct). The one or more consensus algorithms used may be selected or altered to impede an entity from modifying, corrupting, or otherwise altering results of the computation by peer nodes not under the entity's control. Examples of a decentralized tamper-evident data store may include Interplanetary File System, Blockstack, Swarm, or the like. Examples of a decentralized computing platform may include Hyperledger (e.g., Sawtooth, Fabric, or Iroha, or the like), Stellar, Ethereum, EOS, Bitcoin, Corda, Libra, NEO, or Openchain.
The associative array of conditions 220 may include a set of keys, where each key represents an event that may trigger at least one triggerable vertex that would result in a change in norm status, and where a corresponding value of each key includes an events submap. The events submap may include a publisher identifier. As shown by the link 221, the publisher identifier may be used as a reference to the key of the associative array of entities. Alternatively, or in addition, the events submap may include a subject identifier, which may include natural text language to provide a context for the corresponding event.
The associative array of norms 230 may include a set of keys, where each key may represent a norm of the smart contract, which may be associated with as a norm vertex in a graph, norm conditions and consequent norms. In some embodiments, the consequent norms may themselves be associated with their own norm vertices. Each value corresponding to the norm may include a norms submap that includes one or more norm conditions that may be used to trigger the norm by satisfying a norm condition, or by not satisfying the norm condition after satisfying condition expiration threshold associated with the norm. As shown by the link 231, the norm conditions may include a norm identifier that may be used as a reference to a key of the associative array of conditions 220. The norms submap may also include an entity identifier, where the entity identifier may be used as reference to a key of the associative array of entities 210, as shown by the link 232. The norm may also include a condition expiration threshold, which may be represented by the “expiry” field shown in the associative array of norms 230. As discussed above, some embodiments may result in a norm status change or trigger other updates to a vertex if a norm condition is not satisfied but the condition expiration threshold is satisfied. The norm submap may also include a consequences list, where the consequences list may include set of sublists that includes a tail vertex representing a consequent norm that become triggerable, a head vertex of the new norm (which may be the triggered norm), and a label.
In some embodiments, a smart contract state may initially construct the graph list 240 in a first iteration based on the associative array of norms 230 and update the graph list 240 based on a previous iteration of the graph list 240. As described above, the graph list may be in a serialized form, such as a serialized array of norm vertices written in the YAML markup language. As discussed above, the graph list 240 may be a list of graph sublists, where each sublist includes a tail vertex value, a head vertex value, a label associated with the graph edge connecting the tail vertex with the head vertex, a group identifier, and a norm status value. In some embodiments, the norm status may include values such as “satisfied,” “exercised,” “failed,” “active,” “terminal,” “canceled,” “triggerable,” or “untriggerable.” In some embodiments, a norm vertex may be associated with more than one norm status. As shown by link 241, a tail vertex of the graph may be linked to a norm in the associative array of norms 230. Similarly, as shown by the links 242-243, the tail and head vertices of the graph list 240 may be associated with a listed tail norm or head norm in the associative array of norms 230 for a respective norm. Furthermore, as shown by the link 244, the group identifier listed in a graph sublist may also be associated with a value in the associative array of norms 230, such as with a key in the associative array of norms 230.
In some embodiments, a smart contract state may initially construct the knowledge list 250 in a first iteration based on the associative array of norms 230 and update the knowledge list 250 based on smart contract state changes. The knowledge list 250 may be sequentially ordered in time (e.g. a time when a norm status changes, a time when an event is received, or the like). In some embodiments, each entry of the knowledge list 250 may include an identifier “eid,” an event time “etime,” a publisher identifier associated with an event that triggered a norm vertex, the event that triggered the norm vertex. In addition, the knowledge list 250 may include various other data related to the smart contract state change, such as a field “negation” to indicate whether an event is negated, a field “ptime” in ISO8601 format to represent an sub-event time (e.g. for event that require multiple sub-events to trigger a norm vertex), a field “signature” to provide a signature value that allows authentication against the public key held by a publisher for later data authentication operations or data forensics operations. In some embodiments, the knowledge list 250 may include an evidence list, where the evidence list may include a base64 encoded blob, an evidence type containing a string describing the file type of the decoded evidence, and a field for descriptive purposes. In some embodiments, the evidence list may be used for additional safety or verification during transactions.
As described above, some embodiments may efficiently store or update program state data using a set of serialization or deserialization operations. Some embodiments may assign outcome scores to possible outcomes of an update operation, which may then be used to predict future states of a program. Some embodiments may perform operations, such as those described further below, to predict an outcome score using data encoded in a directed graph with greater efficiency or accuracy.
Graph Outcome Determination in Domain-Specific Execution Environment
In some embodiments, outcomes of symbolic AI models (like the technology-based self-executing protocols discussed in this disclosure, expert systems, and others) may be simulated and characterized in various ways that are useful for understanding complex systems. Examples of symbolic AI systems include systems that may determine a set of outputs from a set of inputs using one or more lookup tables, graphs (e.g. a decision tree), logical systems, or other interpretable AI systems (which may include non-interpretable sub-components or be pipelined with non-interpretable models). The data models, norms, or other elements described in this disclosure constitute an example of a symbolic AI model. Some embodiments may use a symbolic AI model (like a set of smart contracts) in order to predict possible outcomes of the model and determine associated probability distributions for the set of possible outcomes (or various population statistics). Features of a symbolic AI model that incorporates elements of data model described in this disclosure may increase the efficiency of smart contract searches. In addition, the use of logical categories (e.g., “right,” “permission,” “obligation”) describing the relationships between conditional statements (or other logical units) of a smart contract may allow the accurate prediction of (or sampling of) outcomes across a population of differently-structured smart contracts without requiring a time-consuming analysis of each of the contexts of individual smart contracts from the population of differently-structured smart contracts. Furthermore, the operations of a symbolic AI model may be used to predict outcomes (e.g., of a smart contract, or call graph of such smart contracts) and may be tracked to logical units (like conditional statements, such as rules of a smart contract). These predicted outcomes may be explainable to an external observer in the context of the terms of the logical units of symbolic AI models, which may be useful in medical fields, legal fields, robotics, dev ops, financial fields, or other fields of industry or research.
In some embodiments, the symbolic AI model may include the use of scores for a single smart contract or a plurality of smart contracts, where the score may represent various values, like a range of movement along a degree of freedom of an industrial robot, an amount of computer memory to be allocated, an amount of processing time that a first entity owes a second entity, an amount to be changed between two entities, a total amount stored by an entity, or the like. A symbolic AI model may include scores of different type. Changes in scores of different type may occur concurrently when modeling an interaction between different entities. For example, a first score type may represent an amount of computer memory to be stored within a first duration and a second score type may represent an amount of computer memory to stored within a second duration that occurs after the first duration. A smart contract may be used to allocate computer memory across two different entities to optimize memory use across the entity domains. Possible outcomes and with respect to memory allocation across the two domains may be simulated. Alternatively, or in addition, exchanges in other computing resources of the same type or different types may be simulated with scores in a symbolic AI model. For example, a symbolic AI model may include a first score and as second score, where the first score may represent an amount of bandwidth available for communication between a first entity or second entity and a third entity, and where the second score may represent an amount of memory available for use by the first or second entity. The outcome of an exchange negotiated via a smart contract between the first and second entity for bandwidth and memory allocation may then be simulated to predict wireless computing resource distribution during operations of a distributed data structure across a wireless network or other computing operations.
In some embodiments, simulating outcomes of may include processing one or more norm vertices representing one or more norms of a smart contract as described in this disclosure. For example, the symbolic AI model may include an object representing a norm vertex, where the object includes a first score representing an amount owed to a first entity and a second score representing an amount that would be automatically transferred to the first entity (e.g., as a down payment). In some embodiments, the symbolic AI model may incorporate the entirety of a smart contract and its associated data model when performing simulations based on the smart contract. For example, a symbolic AI model may include one or more directed graphs of to represent the state of a data model. Alternatively, or in addition, some embodiments may include more data than the smart contract being simulated or less data than the smart contract be simulated.
In some embodiments, the symbolic AI system (a term used interchangeably with symbolic AI model) may process the conditional statements (or other logical units) associated with each of the norms of a smart contract to increase simulation efficiency by extracting only quantitative changes and making simplifying assumptions about score changes. For example, a system may collect the norm conditions and associated outcome subroutines associated with each of a set of norm vertices and extract only the changes in an amount of currency owed as a first score and changes in an amount of currency transferred as a seconds score when incorporating this information into the conditions of the symbolic AI model. In some embodiments, the information reduction may increase computation efficiency by removing information from the analysis of a smart contract determined to be not pertinent to a selected score. Some embodiments simulate outcomes across a plurality of smart contracts using a standardized search and simulation heuristic, and the system described herein may provide a population of scores, where the population of scores may be the plurality of outcome scores determined from a simulation of each of the smart contracts or values computed from the plurality of outcome scores. For example, values determined based on the population of scores may include parameters of a probability distribution of the scores, a total score value, a measure of central tendency (e.g. median score value, mean score value, etc.), or the like.
In some embodiments, the symbolic AI model may be an un-instantiated smart contract or may be a transformation thereof, e.g., approximating the smart contract. For example, as further described below, the system may instantiate a program instance that includes a symbolic AI model based on a selected smart contract that is not yet instantiated. Alternatively, a symbolic AI model may be determined based on an instantiated smart contract. For example, the system may select an instantiated smart contract with a program state that has already changed from its initial program state in order to determine future possible outcomes in the context of the existing changes. The system may then copy or otherwise use a simulated version of the changed program state when simulating the instantiated smart contract. For example, the system may select an instantiated smart contract for simulation with a symbolic AI system and deserialize a directed graph of the instantiated smart contract. The symbolic AI system may copy the deserialized directed graph to generate a simulation of the directed graph, where the nodes of the simulated directed graph are associated with simplified conditional statements that convert quantifiable changes into scores and are stripped of non-quantifiable changes in comparison to the conditional statements of the smart contract.
In addition, or alternatively, the system may perform a lookup operation to select which of the smart contracts to access based on an entity or entity type. For example, the system may compare an entity to the associative arrays of entities corresponding to each of a plurality of smart contracts and select a set of smart contracts based on which of the corresponding arrays of entities include the entity. An entity identifier may be in an array of entities or some other set of entities if an entity type associated with the entity identifier is in the array of entities. For example, if the entity “BLMBRG” has an associated entity type of “trusted publisher,” some embodiments may determine that “BLMBRG” is in the set of entities of a smart contract if the entity type “trusted publisher” is listed in the set of entities. Alternatively, some embodiments may require that the exact entity identifier be listed in a set of entities before determining that the entity identifier in the set of entities. For example, some embodiments may determine that “BLMBRG” is in a set of entities of a smart contract only if “BLMBRG” is one of the elements of the set of entities. Furthermore, in some embodiments, the search may include intermediary entities between two different entities, where intermediary smart contract may be a smart contract (other than the first or second smart contract) that has relationships with both the first and second entities. For example, a search for smart contracts relating a first entity and a second entity may return a set smart contracts that include a first smart contract and a second smart contract, where the array of entities of the first smart contract includes the first entity and an intermediary entity, and where the array of entities of the second smart contract includes the second entity and the intermediary entity.
In some embodiments, an intermediary entity for a first entity and a second entity may be found by determining the intersection of entities between a first set of smart contracts associated with the first entity and a second set of smart contracts associated with the second entity. For example, the system may select a first set of smart contracts from a plurality of smart contracts based on which sets of entities associated with plurality of smart contracts include the first entity. Similarly, the system may select a second set of smart contracts from a plurality of smart contracts based on which sets of entities associated with plurality of smart contracts include the second entity. The system may then determine the intersection of entities by searching through the sets of entities of the first and second set of smart contracts to collect the entities that appear in both the first set and second set and determine that these collected entities are intermediary entities. In some embodiments, as further described below, additional methods are possible to determine a set of smart contracts associating a first entity with a second entity in order to quantify a relationship between the first entity and the second entity.
As discussed in this disclosure, some embodiments may crawl a call graph to select additional smart contracts based on possible relationships between a first entity and a second entity. The call graph may be a privity graph, which may track privity relations between the first entity and entities other than the second entity in order to determine or quantify relations between the first entity and the second entity. if For example, some embodiments may crawl through a privity graph of possible score changes across multiple contracts and determine a quantitative score relationship between a first entity and a second entity based on a first transaction between the first entity and a third entity, a second transaction between the third entity and a fourth entity, a third transaction between the fourth entity and a fifth entity, and a fourth transaction between the fifth entity and the second entity.
In some embodiments, the process 300 includes performing one or more operations indicated by blocks 312, 316, 320, 324, 328, 336, 340, 344, and 350 for each of the respective smart contracts or other programs of the selected set of smart contracts or other programs, as indicated by block 308. As further discussed below, the one or more outputs from executing each of the smart contracts may be used to determine a population of scores of multiple smart contracts. As used herein, the population of scores of multiple smart contracts may represent one or more population metric values calculated from scores of the smart contract. For example, the population of scores of multiple smart contracts may include a measure of central tendency, a measure of dispersion, a kurtosis value, a parameter of a statistical distribution, one or more values of histogram, or the like. Furthermore, in some embodiments, the process 300 may include performing one or more operations in parallel using multiple processor cores, where performing multiple operations in parallel may include performing the multiple operations concurrently. For example, some embodiments may perform the operations of the blocks 312, 316, 320, 324, 328, 336, 340, 344, and 350 for a plurality of smart contracts in parallel by using one or more processors for each of the plurality of smart contracts. By performing operations in parallel, computation times may be significantly reduced.
In some embodiments, the process 300 includes acquiring a set of conditional statements (or other logical units), set of entities, set of indices indexing the conditional statements, or other data associated with the selected smart contract, as indicated by block 312. Each of the set of conditional statements may be associated with an index value and may include or be otherwise associated with a respective set of conditions and a respective set of outcome subroutines, where a computing device may execute the respective set of outcome subroutines in response to an event satisfying the respective set of conditions. In some embodiments, the set of conditional statements may form a network, like a tree structure, with respect to each other. For example, an outcome subroutine of one the conditional statements may include a reference to or otherwise use an index value associated with another conditional statement. In some embodiments, the set of conditional statements and set of indices may be acquired from a data model, where the index values may be or otherwise correspond to the identifiers for norm vertices of a directed graph. For example, the set of conditional statements and set of indices may be acquired from the associative array of norms 230, the associative array of conditions 220, and the graph list 240. Alternatively, the system may acquire the conditional statements and indices from data stored using other data models. For example, the system may acquire the conditional statements from an indexed array of objects, where each object may include a method that can take an event as a parameter, test the event based on a condition of the method, and return a set of values or include a reference to another object of the array. The system may use the indices of the indexed array as the indices of the conditional statements and parse the methods to provide the set of conditional statements.
In some embodiments, the process 300 includes instantiating or otherwise executing a program instance having program state data that includes a symbolic AI model that includes values from the data associated with the selected smart contract, as indicated by block 316. In some embodiments, the symbolic AI model may include graph vertices associated with the set of conditional statements described in this disclosure and may also include directed graph edges connecting the graph vertices. In addition, or alternatively, the symbolic AI model may include a set of tables, decision trees, graphs, or logical systems to provide a predicted value as an output based on one or more inputs corresponding to real or simulated events. For example, the system may traverse the directed graph of a symbolic AI model to determine which nodes of the directed graph to visit based on a decision tree of the symbolic AI model. Furthermore, in some embodiments, the symbolic AI system may be re-instantiated or be modified in real-time in response to a particular event message updating a smart contract being simulated. For example, an instantiated smart contract may be executing and concurrently being simulated by a symbolic AI system. In response to the smart contract receiving an event message, the symbolic AI system may determine a new set of events based on the event message and update its own program state such that its new initial state is based on the smart contract program state after the smart contract program state has been updated by the events of the event message.
In some embodiments, the symbolic AI model may include a graph. In some embodiments, the system may generate a graph list such as the graph list 240 using the methods discussed in this disclosure. In some embodiments, the program instance may be a local version of a selected smart contract and have program state data identical to program state data in the selected smart contract. Alternatively, the program instance may include program data not included in the smart contract or exclude data included in the smart contract. In some embodiments, the graph of the symbolic AI model may include a set of graph vertices and a set of directed graph edges connecting the graph vertices, where each of the graph vertices may be identified by an identifier and corresponds to a conditional statement of a smart contract. In some embodiments, the identifier may be the set of index values associated with the conditional statements of the smart contract. Alternatively, the identifier may be different from the set of index values associated with the conditional statements of the smart contract. For example, the system may choose a set of identifiers that are different from the set of index values to increase system efficiency or reduce memory use.
In some embodiments, the directed graph edges may be structured to provide directional information about the graph vertices of a symbolic AI model. For example, a directed graph edge may be represented as an array of identifier pairs. The first element of each of the identifier pairs may be treated as a tail vertex by the symbolic AI system and the second element of the identifier pairs may be treated as a head vertex by the symbolic AI system. In some embodiments, the selected smart contract may already be in the process of being executed and the program state data of the program instance may include the norm statuses and scores of the smart contract state. For example, the program state data may be copied directly from the state data of a selected smart contract, where the changes effected by the outcome subroutines may be treated as scores.
A smart contract score may represent one of various types of values. For example, a smart contract score may represent a reputation score of an entity in a social network, a cryptocurrency value such as an amount of cryptocurrency, an amount of electrical energy, an amount of computing effort such as Ethereum's Gas, an amount of computing memory, or the like. A smart contract score may represent an objective value associated with an entity, such as an available amount of computing memory associated with the entity. Alternatively, a smart contract score may represent an amount by which a stored value is to be changed, such as a credit amount transferred from a first entity to a second entity.
In some embodiments, a program state may keep track of a plurality of scores. For example, a vertex of a directed graph of a symbolic AI model may include or otherwise be associated with a first score representing an amount of possessed by a first entity, a second score representing an amount owed to or owed by the first entity, a third score representing an amount possessed by a second entity, and a fourth score representing an amount owed to or owed by the second entity. In some embodiments, a conditional statement may be parsed to determine outcome scores. For example, an outcome subroutine associated with a vertex of a graph of the symbolic AI model may include instructions that a first entity is obligated provide 30 cryptocurrency units to a second entity and that the second entity is obligated to send a message to the first entity with an electronic receipt, and the system may determine that an associated score of the first vertex is equal to 30 and also determine that no score value is needed for the sending of the message. As further discussed below, by keeping track of scores and score changes, entire populations of smart contracts may be analyzed with greater accuracy without requiring a deep understanding of the specific terms or entity behaviors of any specific contract.
In some embodiments, a symbolic AI model may include statuses corresponding to each of a set of vertices representing the norms of a smart contract. The symbolic AI model statuses may use the same categories as the norm statuses of a smart contract. Furthermore, the symbolic AI model status for a vertex may be identical to or be otherwise based on the status for the corresponding norm vertex being simulated. For example, if a norm status for a first norm vertex of a smart contract is “triggered—satisfied,” the symbolic AI model status for a first symbolic AI model vertex corresponding to the first norm vertex may also be “triggered—satisfied.” Alternatively, the system may select a different categorical value for a symbolic AI model vertex status that is still based on the corresponding norm status. Similarly, the symbolic AI model may include vertex categories similar to or identical to the logical categories associated with of the set of norm vertices of a smart contract. Furthermore, the symbolic AI model vertex category may be identical to or be otherwise based on the logical for the corresponding norm vertex being simulated. For example, if a logical category for a first norm vertex of a smart contract is “Rights” the symbolic AI model category for a first symbolic AI model vertex (“vertex category”) corresponding to the first norm vertex may also be “Rights.” Alternatively, the system may select a different categorical value for a vertex category that is still based on the corresponding logical category.
In some embodiments, the instantiated program may be a smart contract that may use or otherwise process events. Alternatively, or in addition, the program instance may be a modeling application and not an instance of the selected smart contract itself. For example, a symbolic AI system may be a modeling application that determines the values of a corresponding symbolic AI model based on the conditional statements of a smart contract without requiring that an event message be sent to an API of the modeling application. In some embodiments, the program instance of the symbolic AI system may change program state without performing one or more operations used by the smart contract that the program instance is based on. For example, the program instance of the symbolic AI system may change its program state data without deserializing serialized smart contract data, even if the smart contract that the program instance is based on includes operations to deserialize serialized smart contract data. In some embodiments, the program state data may be stored using a data model similar to that described in this disclosure for
In some embodiments, the process 300 includes performing one or more iterations of the operations indicated by blocks 320, 324, 328, 332, 336, and 340 for each of the respective smart contracts or other programs of the selected set of smart contracts or other programs, as indicated by block 320. Furthermore, in some embodiments, the process 300 may include performing the one or more iterations in parallel using multiple processor cores. For example, some embodiments may include performing multiple iterations of the operations of the blocks 320, 324, 328, 332, 336, or 340 for multiple iterations in parallel using a plurality of processor cores. By performing the multiple iterations of the operations in parallel, computation times may be significantly reduced.
In some embodiments, the system may perform one or more iterations of operations to modify the statuses of a first set of vertices and then update the program state data based on the modified statuses in order to acquire a plurality of outcomes. The program state data or a portion of the program state data may be in a same state at the start each iteration, where two states of program state data are identical if both states have the same set of values. For example, if a first state of program state data is [1,2,3], and if a second state of program state data is [1,2,4], and if the program state data is reverted to [1,2,3], the reverted program state data may be described as being in the first state. In some embodiments, the system may execute the smart contract or smart contract simulation for a pre-determined number of iterations. Alternatively, or in addition, as further recited below, the smart contract or smart contract simulation may be repeatedly executed until a set of iteration stopping criteria are achieved. As further discussed below, the plurality of outcomes corresponding to the plurality of iterations may be used to provide one or more multi-iteration scores usable for decision-support systems and for determining multi-protocol scores.
In some embodiments, the system may modify one or more statuses associated with the vertices of the graph of the symbolic AI model based on a scenario and update the program state data based on the modified statuses, as indicated by block 328. In some embodiments, the scenario may be a set of inputs based on events. For example, a scenario may include simulated events or simulated event messages that may be testable by the conditions of a conditional statement. In response, a first vertex of the program instance may compare the simulated event to a condition and determine that a second vertex of the symbolic AI model of the should be activated. For example, an input may include an event “entity A transmitted data 0x104ABC to entity C,” which may satisfy a condition and change a status associated with a first vertex associated with the conditional statement to “satisfied.” As discussed below, the system may then update the symbolic AI model based on the status change by activating an adjacent vertex to the first vertex.
Alternatively, or in addition, an input may include a message to change a program state without including an event that satisfies the norm conditions associated with the norm. For example, the input may include direct instructions interpretable by a symbolic AI system to set a vertex status to indicate that the corresponding vertex is triggered and direct which of a set of outcome subroutines to execute. The system may then update the symbolic AI model by activating one or more adjacent vertices described by the subset of outcome subroutines to execute.
In some embodiments, the scenario may include a single input. Alternatively, the scenario may include a sequence of inputs. For example, the scenario may include a first event, second event, and third event in sequential order. In some embodiments, the set of events may be generated using a Monte Carlo simulator. Some embodiments may randomly determine subsequent states from an initial state based on one or more probability distributions associated with each state of a set of possible subsequent states with respect to a previous state, where the probability distributions may be based on scores and logical categories associated with the set of possible states. For example, the program state may be in a state where only two subsequent possible states are possible, where the first subsequent possible state includes triggering a rights norm and a second subsequent possible state includes triggering an obligations norm.
In some embodiments, one or more inputs of a scenario may be determined using a decision tree. In some embodiments, a decision tree may be used to provide a set of decision based on scores, logical categories, statuses, and other factors associated with the active vertices of a simulated smart contract state. For example, a symbolic AI system may determine that the two possible states for a smart contract may result from either exercising a first rights norm or exercising of a second rights norm. A decision tree may be used to compare the logical categories, the scores associated with each norm, and the other information related to the active norms to determine which rights norm an entity would be most likely to exercise. In some embodiments, the symbolic AI system may compare a first score associated with a possible state represented by a first tree node with a second score of a different possible state represented by a second tree node. In response to the first score being greater than the second score, the symbolic AI system may determine a simulated input that will result in the future state represented by the first tree node. Furthermore, in some embodiments, the decision tree may incorporate probability distributions or other decision-influencing factors to more accurately simulate real-world scenarios.
Alternatively, or in addition, some embodiments may include a Monte Carlo Tree Search (MCTS) method to generate a random sequence of events based on a set of possible events and a probability distribution by which the events may occur. The operations of the simulation may be made more efficient by selecting events that known to satisfy at least one condition of the set of conditional statements of the smart contract being simulated. In some embodiments, a symbolic AI system may determine a set of events for a smart contract simulation by determining a first simulated input based on a set of weighting values assigned to vertices of a graph of a symbolic AI model associated with norms of the smart contract. In some embodiments, the system may further determine a simulated input based on a count of the number of iterations of the simulation performed so far.
The system may then update the symbolic AI model based on the first simulated input, advancing the symbolic AI model to a second state. For example, after changing the status of a first vertex associated with an obligations norm from “unrealized” to “failed,” the symbolic AI model may then activate a first adjacent vertex representing a rights norm and a second adjacent vertex representing an prohibitions norm, where both adjacent vertices are adjacent to the first vertex. The symbolic AI system may then determine a second simulated input, wherein the second simulated input may be selected based on weighting value corresponding to each of the first adjacent vertex and second adjacent vertex, where the weighting value may be a score of the smart contract. For example, the weighting value of the first adjacent vertex may be 2/4 and the weighting value of the second adjacent vertex may be 1/6. Some embodiments may then update the symbolic AI model when it is in the second state based on the second simulated input in order to advance the second model to a terminal state, where a terminal state is one that satisfies a terminal state criterion. Once in a terminal state, the symbolic AI system may update the weighting values associated with the symbolic AI model before performing another iteration of the simulation.
Various terminal state criteria may be used. For example, a terminal state criterion may be that there is no further state change possible. Alternatively, a terminal state criterion may be that the smart contract is cancelled. The system may then update each of the weighting values associated with each of the nodes after reaching a terminal state before proceeding to perform another iteration. In some embodiments, the symbolic AI system may set a status of a vertex to “failed” to simulate the outcomes of a first entity failing to transfer a score (e.g. failure to pay) a second entity.
In some embodiments, the determination of an input may be based on the type of conditional statement being triggered. As further discussed below, one or more of the conditional statements may be non-exclusively classified as one or more types of norms. Example of norm types include rights norms, obligations norms, or prohibition norms. As further discussed below, norm types may also include associations as being part of a pattern, such as a permission pattern. For example, a vertex may include or be otherwise associated with the label “consent or request.” By determining activities based on logical categories associated with the conditional statements instead of specific events, predictive modeling may be performed using globalized behavior rules without interpreting each of the globalized behavior rules for each specific contract. For example, a sequence of event may be generated a based on a first probability distribution that approximates an obligation of a first entity as having a 95% chance of being fulfilled and a 5% chance of being denied and a second probability distribution that approximates that a second entity has a 10% chance of cancelling a smart contract before the first entity exercises a right to cure the failure to satisfy the obligation. Using these rules, population scores associated with the population of smart contracts between a first entity and a second entity that consist of obligations norms to pay, rights norms to cure, and rights norms to cancel may be determined without regards to the specific structure of individual smart contracts in the population of smart contracts.
The system may then update each of the smart contract instances associated with a changed norm status, as discussed further below. Furthermore, the system may then update the respective associative array of conditions corresponding to the set of smart contracts. In some embodiments, an associative array of conditions may include only a subset of norm conditions associated with a smart contract, where each the subset of norm conditions is associated with a triggerable vertex of the smart contract. In some embodiments, the system may first deduplicate the norm conditions before performing a lookup operation to increase performance efficiency. For example, after determining that an event has occurred, some embodiments may search through a deduplicated array of norm conditions. For each norm condition that the event would trigger, the system may then update the one or more smart contracts associated with the norm condition in the deduplicated array of norm conditions.
Some embodiments may obtain a sequence of inputs instead of a single input. In some embodiments, the system may use a neural network to generate the sequence of inputs. In some embodiments, the neural network may determine a state value s based on the program state data and provide a vector of probabilities associated with for each of a set of possible changes in the program state. The neural network may also determine a state value to estimate the expected value of the program state after system applies the scenario to the program. In some embodiments, the neural network may use a MCTS algorithm to traverse a tree representing possible future states of the smart contract from a root state. The system may determine a next possible state s+1 for each state s by selecting a state with a low visit count, high predicted state value, and high probability of selection. The parameters (e.g. weights, biases, etc.) of the neural network making the state value determination may be represented by θ. After each iteration ending in a terminal state, the system may adjust the values θ to increase the accuracy of the neural network's predicted state value in comparison to the actual state value assessed whenever a terminal state is reached. Furthermore, a symbolic AI model may have a total score value, and the system may update the total score value based on the state value.
In some embodiments, the process 300 includes determining an outcome score based on the updated program state data, as indicated by block 336. In some embodiments, as stated in this disclosure, a set of scores may be associated with one or more of the outcome states. For example, an outcome of a first norm may include a transfer of currency values from a first entity to a second entity. The symbolic AI system may record this score and combine it with other scores in the same iteration in order to determine a net score for that score type. For example, the symbolic AI system may record each currency change based on inputs and outcomes in order to determine a net currency change, where a score of the smart contract may be the net currency change. Alternatively, or in addition, the symbolic AI system may record scores across different iterations to determine a multi-iteration score, as described further below. Example outcome scores may include a net amount of currency exchanged, a net amount of computing resources consumed, a change in the total cryptocurrency balance for an entity, or the like.
The process 300 may execute a number of iterations of smart contract state change simulations to determine possible outcomes and outcome scores. In some embodiments, there may be one or more criteria to determine if an additional iteration is needed, as indicated by block 340. In some embodiments, the one or more criteria may include whether or not a pre-determined number of iterations of simulations have been executed. For example, some embodiments may determine that additional iterations are needed if the total number of executed iterations is less than an iteration threshold, where the iteration threshold may be greater than five iterations, greater than ten iterations, greater than 100 iterations, greater than 1000 iteration, greater than one million iterations, greater than one billion iterations, or the like. Alternatively, or in addition, the one or more criteria may include determining whether a specific outcome occurs. For example, the one or more criteria may include determining whether the outcome score is less than zero after a terminal state is reached. If the additional iterations are needed, operations of the process 300 may return to block 320. Otherwise, operations of the process 300 may proceed to block 344.
In some embodiments, the process 300 includes determining a multi-iteration score based on the outcome scores of executed iterations, as indicated by block 344. The multi-iteration score may be one of various types of scores and may include values such as a net change in score across multiple iterations, a probability distribution parameter, a measure of central tendency across multiple iterations, a measure of dispersion, or a measure of kurtosis. For example, the system may use a first outcome score from a first iteration, a second outcome score from a second iteration, or additional outcome scores from additional iterations to determine an average outcome score. The system may determine additional multi-iteration scores in the form of probability distribution parameters to determine a probability distribution. As used herein, a measure of kurtosis value may be correlated with a ratio of a first value and a second value, wherein the first value is based on a measure of central tendency, and wherein the second value is based on a measure of dispersion. For example, the measure of kurtosis may equal to μ4/σ4, where μ may be a fourth central moment of a probability distribution and σ may be a standard deviation of the probability distribution.
In some embodiments, the multi-iteration score may be used to provide one or more predictions using Bayesian inference methods. In some embodiments, the multi-iteration score may be used to generate a probability distribution for the probability that a particular event or event type occurred based on a score, such as a change in currency value or an amount of computing resources consumed. For example, the system may calculate a mean average cryptocurrency amount determined across multiple iterations as a first multi-iteration score and a standard deviation of the cryptocurrency amount as the second multi-iteration score while tracking the number of payment delays associated with the respective cryptocurrency amounts. The system may then use the first and second multi-iteration scores to generate a gaussian distribution, where the system may use the gaussian distribution to perform Bayesian inferences in order to determine a probability that a payment delay occurred after obtaining the value of a new cryptocurrency amount.
In some embodiments, the multi-iteration score may be a weight, bias, or other parameter of a neural network. For example, some embodiments may use a set of multi-iteration scores as weights of a neural network, where the training inputs of the neural network may be outcome scores and the training outputs of the neural network may be events, indicators representing activated outcome subroutines, or activated patterns. Once trained, the neural network may determine the probability of events, triggered conditional statements, or triggered patterns based on observed scores. In some embodiments, the parameters of the neural network may be transferred to other neural networks for further training. For example, a first neural network may be trained using the outcome scores as inputs and sets of events as outputs, and the weights and biases of the training may be transmitted to a second neural network for further training. The second neural network may then be used to indicate whether a particular event had a sufficiently high possibility of occurring based on a score or score change. In addition, the multi-iteration score may include outputs of a convolutional neural network, which may be used to determine behavior patterns across multiple smart contracts.
In some embodiments, the symbolic AI system may use a fuzzy logic method to predict the occurrence of an event based on the outcomes of a smart contract. A fuzzy logic method may include fuzzifying inputs by using one or more membership functions to determine a set of scalar values for each of a set of inputs, where the set of scalar values indicate the degree of membership of the inputs of a set of labels for each of the inputs of a smart contract being simulated by a symbolic AI system. For example, the system may use a membership function to determine a percentage values between 0% and 100% for a set of labels such as “profitable,” “risky,” or the like. The percentage values may indicate, for each of the smart contracts, a degree of membership to each of the labels. The symbolic AI system may then determine an fuzzified outcome score based on the set of fuzzified data by first using a set of rules in combination with an inference engine to determines the degree of match associated with the fuzzy input and determine which of the set of rules to implement. As used herein, an inference engine may be a system that applies a set of pre-defined rules. For example, an inference engine may include a set of “if-then” expressions that provided responses to particular inputs. By using the inference engine in combination with the set of rules, the fuzzified outcome score may provide an indication of a broader label for the smart contract, such as “unconventional,” “risk too high,” or the like. In some embodiments, the symbolic AI system may defuzzify the fuzzified outcome score using various methods such as centroid of area method, bisector of area method, mean of maximum method, or the like. The defuzzifying process may result in a defuzzified outcome score that may also be used to determine a label.
In some embodiments, each of the scenarios may have an associated scenario weight, where the associated scenario may be a numeric value representing a normalized or nonnormalized probability of occurrence. For example, a smart contract may be processed based one of three possible scenarios, where the first scenario may have a weighting value equal to 0.5, the second scenario may have a weighting value equal to 0.35, and the third scenario may have a weighting value equal to 0.15. The system may use the associated scenario weights when determining a multi-iteration score. For example, if the first, second, and third scenarios results in allocating, respectively, 100, −10, or −100 computing resource units to a first entity, the system may determine that the expectation resource units allocated to the first entity is equal to 31.5 computing resource units and use the expectation resource units as the allocation value. While the above described using a scalar value as a weighting value, some embodiments may instead use a probability distribution as an associated scenario weight for each of the scenarios and determine the weighting value.
In some embodiments, the system may determine if data from an additional smart contract is to be processed, as indicated by block 350. As discussed in this disclosure, the process 300 may execute a number of simulations of different smart contracts to simulate possible outcomes and score changes. In some embodiments, each of the set of selected smart contracts may be simulated using a symbolic AI simulator. Furthermore, each of the set smart contracts may use the same set of weights/probability values to determine unique scenarios. For example, using the same set of weights corresponding to different combinations of available vertices, the system may determine a first scenario for a first symbolic AI model and a second scenario for a second symbolic AI model, where the first and second symbolic AI models have directed graphs that are different from each other. In some embodiments, the same weights may be used because the plurality of symbolic AI models may include vertices based on the same set of statuses and same set of logical categories. If the additional iterations are needed, operations of the process 300 may return to block 308. Otherwise, operations of the process 300 may proceed to block 354.
In some embodiments, the process 300 includes determining a multi-protocol score based on the outcome scores across multiple smart contracts, as indicated by block 354. A multi-protocol score may be any score that is determined based on a plurality of outcomes from simulating different smart contracts, where the plurality of outcomes may include either or both multi-iteration scores or scores determined after a single iteration. In some embodiments, the multi-protocol score may be determined by determining a population of scores associated with a given entity. For example, a population of scores may be a population of expected income across a population of 500 instantiated smart contracts. The multi-protocol score may be a total income value, average income value, kurtosis income value, or the like.
In some embodiments, one or more methods to determine a multi-iteration score may be used to determine a multi-protocol score. For example, use of fuzzy logic, Bayesian inference, or neural networks may be used to predict multi-protocol scores. For example, some embodiments may use a first set of multi-iteration scores from a plurality of smart contract simulations as inputs and a second set of multi-iteration scores from the same plurality of smart contract simulations as outputs when training a neural network, where a set of multi-protocol score may be one or more the parameters of the trained neural network. For example, some embodiments may include a neural network trained to predict the probability that a specific type of smart contract was used based on multi-iteration scores such as an average payment duration and an average payment amount.
In some embodiments, multiple multi-protocol scores may be used to determine risk between a first entity and a second entity. For example, operations of the process 300 may be performed to determine a list of smart contracts shared by a first entity and a second entity and predict possible risks to the first entity in scenarios resulting from the incapability of the second entity to fulfill one or more norms in the list of smart contracts. In some embodiments, the risk posed to a first entity by a second entity may include considerations for intermediate relationships. For example, a first entity may be owed multiple amounts from a plurality of entities other than a second entity, and a second entity may owe multiple amounts to the plurality of entities. In some embodiments, a risk associated with the total amount of a score value to be collected by the first entity from the plurality of entities may be assessed based on the risk of the second entity failing to fulfill one or more obligations to transfer score values to one or more of the plurality of entities. While the relationship between the first entity and the second entity may be difficult to determine using conventional smart contract systems if no explicit privity relations are listed in the smart contracts, the symbolic AI models described in this disclosure allow these relationships to be determined by searching through entity lists or crawling through one or more privity graphs.
Box 510 includes a directed graph representing a smart contract state (or simulation of the smart contract state) after an event satisfying a norm condition of the obligation norm represented by the norm vertex 511. As shown in box 510, after a determination that the norm condition P associated with the norm vertex 511 is satisfied by an event (indicated by the directed graph edge 512), the system may generate an adjacent vertex 513 indicating that the norm vertex 511 is satisfied, where a norm status of the adjacent vertex 513 may be set as “terminal” to indicate that the adjacent vertex is terminal. In some embodiments, a determination that the state of the smart contract or simulation thereof is terminal may be made if a vertex of the smart contract or simulation thereof indicated to be terminal. In some embodiments, instead of generating the adjacent vertex 513, the system may update a norm status associated with the norm vertex 511 to indicate that the norm vertex 511 is satisfied. For example, the system may update a norm vertex associated with the obligation norm by setting a norm status associated with the norm vertex to “satisfied,” “terminal,” or some other indicator that the obligation norm has been satisfied by an event. In some embodiments, updating a norm vertex associated with the obligation norm may be represented by the statement 6, where
represents a result of the norm condition associated with the obligation norm “OP” being satisfied, and S represents the generation of a norm vertex indicating that the conditions of the obligation norm have been satisfied:
As shown in box 520, an norm condition P may end up not satisfying a norm condition associated with the norm vertex 521 after satisfying a condition expiration threshold, where the norm vertex 521 is associated with an obligation norm. In response, the system may update the norm vertex 521 by setting a norm status associated with the norm vertex 521 to “failed” or some other indicator that the norm condition associated with the norm vertex 521 has been not satisfied. For example, an event may indicate that a condition expiration threshold has been satisfied without an obligation norm condition being satisfied. In response, the system may generate or otherwise set as triggerable the set of consequent norms associated with adjacent vertices 523, where the relationship between a failure to satisfy a norm condition P of the norm vertex 521 and the adjacent vertices 523 is indicated by the directed graph edges 522. In some embodiments, the generation of the adjacent vertices may be represented by the statement 7, where
indicates that the instructions to the right of the symbol
are to be performed if a norm condition “OP” is not satisfied, and the instructions represented by the symbolic combination ΛiXiQi represents the generation or activation of the consequent norms that result from the failure of OP:
In some embodiments, in response to an event satisfying a norm condition of a rights norm, the system may update a norm vertex associated with the rights norm by setting a norm status associated with the norm vertex to “exercised” or some other indicator that the rights norm has been triggered based on an event. For example, as shown in box 530, in response to an event satisfying a norm condition associated a rights norm represented by the norm vertex 531, the system may update a norm vertex associated with the norm vertex 531 by setting a norm status associated with the norm vertex 531 to “exercised” or some other indicator that the rights norm has been exercised. In response, the system may generate or otherwise set as triggerable the set of consequent norms associated with adjacent vertices 533, where the relationship between satisfying a norm condition P associated with the norm vertex 531 and the set of consequent norms associated with adjacent vertices 533 is indicated by the directed graph edges 532. Furthermore, in some embodiments, a rights norm may be contrasted with an obligation norm by allowing a rights norm to remain triggerable after triggering. This may be implemented by further generating or otherwise setting as triggerable the rights norm associated with the rights norm vertex 534. In some embodiments, a rights norm may expire after use. For example, some embodiments may not generate the rights norm vertex 534 after triggering the norm vertex 531. In some embodiments, the operation described above may be represented by statement 8 below, where the result of triggering a rights norm RP1 by satisfying the norm condition P may result in a conjunction of newly-triggerable consequent norms ΛiXiQi and a rights norm RP2 that is identical to the rights norm RP1, where Λ represents a mathematical conjunctive operation:
In some embodiments, in response to an event satisfying the norm condition of a “prohibition” norm, the system may update a norm vertex associated with the “prohibition” norm by setting a norm status associated with the norm vertex to “violated” or some other indicator that the “prohibitions” norm has been triggered based on an event. For example, as shown in box 550, an event may satisfy a norm condition P associated with the prohibition norm represented by a norm vertex 551. In response, the system may update the norm vertex 551 by setting a norm status associated with the norm vertex 551 to “violated” or some other indicator that the associated prohibitions norm condition has been satisfied. In response, the system may generate or otherwise set as triggerable the set of consequent norms associated with adjacent vertices 553, where the relationship between satisfying a norm condition P associated with the norm vertex 551 and the set of consequent norms associated with adjacent vertices 553 is indicated by the directed graph edges 552. Furthermore, in some embodiments, a prohibitions norm may be contrasted with an obligation norm by allowing a prohibitions norm to survive triggering. In addition, in some embodiments, triggering a prohibitions norm may result in the system decreasing a value representing the state of the smart contract. This may be implemented by further generating or otherwise setting as triggerable the prohibitions norm associated with the prohibitions norm vertex 554 after triggering the norm vertex 551. In some embodiments, the operation described above may be represented by statement 9 below, where the result of triggering a prohibition norm PP1 by satisfying the norm condition P may result in a conjunction of newly-triggerable consequent norms ΛiXiQi and a prohibition norm PP2 that is identical to the prohibition norm PP1, where Λ represents a mathematical conjunctive operation:
may indicate that the event which triggers the norm XP occurs when the norm condition P is either satisfied or failed, ΛiXiQi represents the set of consequent norms that are set to be triggerable based on the event triggering XP, and XjUj may represent the set of consequent norms that cancelled based on event triggering XP:
As shown by statement 10 above, one or more norms may be cancelled. In some embodiments, a cancellation may be implemented as an inactive graph edge between the norm XP and the norms XjUj, where the graph edge representing the conditional relationship between the norm XP and the norms XjUj are directed towards the norm XP. In some embodiments, the cancellation of a norm may be implemented by setting an indicator to indicate that a norm or condition associated with the cancelled norm is no longer triggerable.
The directed graph 610 may represent a state of a start contract and may include the first vertex 611, second vertex 613, third vertex 617, and fourth vertex 619, each of which are associated with a norm of a smart contract. The directed graph 610 also depicts a mutual cancellation relationship between the norm associated with the second vertex 613 and the third vertex 617 represented by the XQ1-XQ2 graph edge 614, where a mutual cancellation relationship of a pair of norm vertices may include a cancellation of one norm vertex of the pair upon triggering of the other norm vertex of the pair. The directed graph 610 also depicts a unidirectional cancellation relationship between the norm associated with the fourth vertex 619 and the third vertex 617 as represented by the XP2-XQ2 graph edge 618. In some embodiments, satisfying or otherwise triggering the norm associated with the third vertex 617 may instantiate the RZ-XQ2 graph edge 618 and cancel the fourth vertex 619. In some embodiments, each of vertices and graph edges shown in
In some embodiments, the state represented by the directed graph 610 may advance to the state represented by the directed graph 620. The state represented by the directed graph 620 may be achieved by triggering the norm associated with the second vertex 613, which may result in the cancellation of the norm associated with the third vertex 617. Furthermore, as illustrated by the directed graph 620, triggering the norm associated with the second vertex 613 may also result in the activation of fifth vertex 621 and sixth vertex 623. In addition, triggering the norm associated with the third vertex 617 may result in the cancellation of the norm associated with the fourth vertex 619. Furthermore, as illustrated by the directed graph 630, triggering the norm associated with the third vertex 617 may also result in the activation of a seventh vertex 631 and eighth vertex 633. Each of these triggering behaviors may be implemented directly by a smart contract.
In some embodiments, the triggering relationship described in this disclosure may be modeled using a symbolic AI system that may keep track of any scores associated with events that trigger the norms and the outcomes of triggering the norms. For example, a first probability value may be assigned to the state represented by the directed graph 620 and a second probability value may be assigned to the state represented by the directed graph 630 during a simulation of the smart contract. The symbolic AI system may use the first and second probability values to advance the state represented by either the directed graph 620 or the directed graph 630 over multiple iterations to compute a multi-iteration score using the methods described in this disclosure. For example, if the first probability value is 20% and the second probability value is 80%, and a first score represented by the directed graph 620 is equal to 100 cryptocurrency units and a second score represented by the directed graph 630 is equal to 1000 cryptocurrency units, a multi-iteration score may be equal to 820 cryptocurrency units.
The right column of table 600 includes a directed graph 650, which may represent an initial state of a smart contract (or simulation thereof). The right column of table 600 also includes a directed graph 660 that represents a first possible outcome state of the initial state and a directed graph 670 that represents a subsequent possible outcome state of the first possible outcome state. The initial state represented by the directed graph 650 may include a permissive condition of a permission norm, where satisfying a permissive condition may result in the activation of one or more norms. For example, after being activated, a rights norm RP may include a set of permissions {RVk} that are triggered after satisfying an norm condition associated with the rights norm RP, where the rights norm RP may also be described as a permission norm. Triggering the set of permissions {RVk} may either set the norm XP to be triggerable or otherwise prevent an outcome subroutine of the norm XP from being executed until the set of permissions {RVk} are triggered. This relationship may be represented by statement 11 below, where XP may represent an obligations norm, RVk represents the permissions that must be triggered before XP may be triggered,
may indicate that the event which triggers the norm XP occurs when the norm condition P is either satisfied or failed, ΛiXiQi represents the set of consequent norms that are set to be triggerable based on the event triggering XP after the permissions RVk are triggered, and XjUj may represent the set of consequent norms that cancelled based on event triggering XP after the permissions RVk are triggered:
As shown by statement 11 above, XP may be set to be triggerable upon triggering of the permission RVk. Triggering XP after the permissions RVk are triggered results in activation of the consequent norms ΛiXiQi and cancels the norms XjUj. In some embodiments, the conditions needed to trigger permissions may be activated in conjunction with rights norms dependent on the permissions, and thus XP and RVk may be activated as a result of triggering the same triggered norm. In some embodiments, permission behavior may be performed by a smart contract or a simulation thereof by modifying a first status of a first vertex and a second status of a second vertex to indicate that the first and second vertices are triggered, where the first vertex may represent a first rights norm such as XP and the second vertex may represent a permission norm such as a norm having outcome permissions RVk. The smart contract, or a simulation thereof, may trigger a third vertex that is adjacent to the first vertex and the second vertex such as a vertex in ΛiXiQi in response to the first and second statuses now being triggered.
The directed graph 650 may include a first vertex 651, second vertex 653, third vertex 657, and fourth vertex 659. The directed graph 650 also depicts a mutual cancellation relationship between the norm associated with the second vertex 653 and the third vertex 657 represented by the XQ1-XQ2 graph edge 654. The directed graph 650 also depicts a permission relationship between the norm associated with the fourth vertex 659 and the third vertex 657 as represented by the RZ-XQ2 graph edge 658, where the fourth vertex 659 may include or otherwise be associated with permission conditions that must be satisfied in order to trigger the third vertex 657. In some embodiments, satisfying or otherwise triggering the norm associated with the fourth vertex 659 may instantiate the RZ-XQ2 graph edge 658 and allow the outcome subroutines of the third vertex 657 to be executed.
In some embodiments, the program state represented by the directed graph 650 may produce an outcome state represented by the directed graph 660. The outcome state represented by the directed graph 660 may be achieved by satisfying a norm condition associated with the fourth vertex 659. In some embodiments, after the XQ1-XQ2 graph edge 654 becomes instantiated, an event satisfying a norm condition associated with the third vertex 657 may result in the program state represented by the directed graph 670. The directed graph 670 may represent a program state where the norm associated with the third vertex 657 is triggered, resulting in the activation of additional norms associated with the fifth vertex 671 and sixth vertex 673.
In some embodiments, a symbolic AI system may be used to generate a scenario that includes a sequence of inputs having a first input and a second input. The first input may advance the state represented by the directed graph 650 to the state represented by the directed graph 660 and the second input may advance the state represented by the directed graph 660 to the state represented by the directed graph 670. The sequence of inputs may be determined using any of the methods described in this disclosure. For example, the sequence of inputs may be determined using a Monte Carlo method, a neural network, or the like.
The set of directed graphs 720 may represent a first outcome state that may result from the program state represented by the set of directed graphs 710, where each of the obligation norms represented by the three vertices 711-713 are satisfied. In some embodiments, a smart contract simulation system such as a symbolic AI system may assign a probability value to the possibility the state represented by the set of directed graphs 710 is advanced to the outcome state represented by the set of directed graphs 720. For example, a symbolic AI system may assign a probability for the outcome state represented by the set of directed graphs 720 to be equal to 82% when starting from the state represented by the set of directed graphs 710. The symbolic AI system may then perform a set of simulations based on this probability value using a Monte Carlo simulator.
The set of directed graphs 730 may represent a second outcome state that may result from the program state represented by the set of directed graphs 710, where the first obligation is not satisfied and the time has exceeded a condition expiration threshold associated with the first vertex 711. As shown in the set of directed graphs 730, a failure to meet the first obligation represented by the first vertex 711 may result in a system generating or otherwise activating norms associated with a fourth vertex 721 and a fifth vertex 722. In some embodiments, the norm associated with the fourth vertex 721 may represent a first entity's right to cure the payment failure and the norm associated with the fifth vertex 722 may represent a second entity's right to terminate the smart contract. The bidirectional graph edge 723 indicates that triggering one of the pair of vertices 721-722 will cancel or otherwise render as inactive the other of the pair of vertices 721, which may indicate that curing a failed obligation and terminating the smart contract may be mutually exclusive outcomes. In some embodiments, a symbolic AI system (or other modeling system) may assign a probability value to the possibility the state represented by the set of directed graphs 710 is advanced to the outcome state represented by the set of directed graphs 720. For example, the symbolic AI system may assign a probability for the outcome state represented by the set of directed graphs 720 to be equal to 6% when performing a simulation based on the smart contract program state represented by the set of directed graphs 710.
In some embodiments, the state represented by the set of directed graphs 730 may be advanced to the state represented by a set of directed graphs 740. In some embodiments, the state represented by a set of directed graphs 740 may be an outcome state after the norm associated with the fourth vertex 721 is triggered. As shown in the set of directed graphs 740, triggering the norm associated with the fourth vertex 721 may result in cancelling the norm associated with fifth vertex 722. In some embodiments, a symbolic AI system may use probability value representing the probability of the state represented by the set of directed graphs 730 advancing to the state represented by a set of directed graphs 740. For example, a symbolic AI system may use 50% as the probability that the state represented by the set of directed graphs 730 advances to the state represented by a set of directed graphs 740. If the probability of the state represented by the set of directed graphs 720 advancing to the state represented by a set of directed graphs 730 is equal to 6%, this would mean that the probability of the state represented by the set of directed graphs 710 advancing to the state represented by a set of directed graphs 740 is equal to 3% by applying the multiplication rule for the probability of independent events.
In some embodiments, the state represented by the set of directed graphs 730 may be advanced to the state represented by a set of directed graphs 750. In some embodiments, the state represented by a set of directed graphs 750 may be an outcome state after the norm associated with the fifth vertex 722 is triggered. As shown in the set of directed graphs 750, triggering the norm associated with the fifth vertex 722 may result in cancelling the norm associated with second vertex 712, third vertex 713, and fourth vertex 721. In some embodiments, a symbolic AI system may assign a probability value to the possibility of a smart contract state being in the outcome state represented by the set of directed graphs 750 when starting from the program state represented by the set of directed graphs 730. In some embodiments, the probability values associated with each state may be updated after each iteration in a set of simulated iterations using one or more of the methods in this disclosure. For example, some embodiments may apply a MCTS method to explore the program states represented by the sets of directed graphs 710, 720, 730, and 740 across multiple iterations while keeping track of scores for each iteration in order to determine outcome scores for each iteration and multi-iteration scores.
In some embodiments, the state represented by the set of directed graphs 810 may be advanced to the state represented by a set of directed graphs 820. In some embodiments, the state represented by a set of directed graphs 820 may be an outcome state after the norm associated with the fifth vertex 822 is triggered. As shown in the set of directed graphs 820, triggering the norm associated with the fifth vertex 822 may result in cancelling the norm associated with sixth vertex 823. In some embodiments, the fifth vertex 822 may represent a first entity's right to terminate the order and obtain a refund. This outcome may be represented by the eighth vertex 831, which may represent an obligation norm indicating that the second entity has an obligation to pay the first entity, and that this obligation may either be satisfied or failed, as indicated by vertices 841 and 842, respectively.
In some embodiments, the state represented by the set of directed graphs 810 may be advanced to the state represented by a set of directed graphs 830. In some embodiments, the state represented by a set of directed graphs 820 may be an outcome state after the norm associated with the sixth vertex 823 is triggered. As shown in the set of directed graphs 830, triggering the norm associated with the sixth vertex 823 may result in cancelling the norm associated with sixth vertex 823. In some embodiments, the sixth vertex 823 may represent a first entity's right to cure the failure to satisfy the norm represented by the second vertex 812. This outcome may be represented by the ninth vertex 832, which may represent an obligation norm indicating that the second entity has an obligation to deliver an asset to the first entity, and that this obligation may either be satisfied or failed, as indicated by vertices 843 and 844, respectively.
In some embodiments, a symbolic AI system may assign a probability value to the possibility of a smart contract state being in the outcome state represented by the set of directed graphs 820 or set of directed graphs 830 when starting from the program state represented by the set of directed graphs 810. For example, a symbolic AI system may determine that the probability that the outcome state represented by the set of directed graphs 820 is equal to 40%. Similarly, the symbolic AI system may determine that the probability that the outcome state represented by the set of directed graphs 830 is equal to 60%. In some embodiments, the symbolic AI system may use a Bayesian inference to determine if an obligation norm was failed was failed based on a probability distribution computed from the scores associated with program states such as those states represented by the sets of directed graphs 820 or 830. For example, the symbolic AI system may acquire a new score value and, based on the score value, predict whether an obligation represented by the second vertex 812 was failed.
In some embodiments, the state represented by the set of directed graphs 910 may be advanced to the state represented by a set of directed graphs 920. In some embodiments, the state represented by a set of directed graphs 920 may be an outcome state after the norm associated with the sixth vertex 923 is triggered. In some embodiments, the norm associated with the sixth vertex 923 may represent a first entity's right to cure the payment failure, and thus triggering the rights norm associated with the sixth vertex 923 may represent a first entity's right to cure the failure. As indicated by the satisfaction vertex 931, curing the payment failure may end all outstanding obligations of the smart contract.
In some embodiments, the state represented by the set of directed graphs 910 may be advanced to the state represented by a set of directed graphs 930. In some embodiments, the state represented by a set of directed graphs 930 may be an outcome state after the norm associated with the seventh vertex 924 is triggered. In some embodiments, the norm associated with the seventh vertex 924 may represent a second entity's right to declare a breach, and thus triggering the rights norm associated with the seventh vertex 924 may represent a second entity's declaration of contract breach. This may result in the activation of the failure vertex 932, which may include outcome subroutines that sends a message indicating that the smart contract is in breach to a third party or sends instructions to an API of another application.
In some embodiments, the state represented by the directed graph 1010 may be advanced to the state represented by the directed graph 1020. In some embodiments, the state represented by the directed graph 1020 may be an outcome state after the norm associated with the second vertex 1012 is triggered. In some embodiments, the norm associated with the second vertex 1012 may represent a right to cure the failure to satisfy the norm condition associated with the first vertex 1011. As indicated by the directed graph 1020, exercising the rights norm associated with the second vertex 1012 may satisfy the norm and activate the vertex 1023, which may indicate that the rights norm associated with the second vertex 1012 has been satisfied.
In some embodiments, the state represented by the directed graph 1010 may be advanced to the state represented by the directed graph 1030. In some embodiments, the state represented by the directed graph 1030 may be an outcome state after the norm associated with the third vertex 1013 is triggered. In some embodiments, the rights norm associated with the third vertex 1013 may represent a right to accelerate payment. Triggering the rights norm associated with the third vertex 1013 may cancel the rights norm associated with the second vertex 1012. In addition, triggering the rights norm associated with the third vertex 1013 may also cancel the obligation norms associated with the vertices 1014-1015. Triggering the rights norm associated with the third vertex 1013 may cause the system to activate a new obligation norm associated with the fourth vertex 1031. In some embodiments, the new obligation norm may include norm conditions to determine whether a first entity transmits a payment amount to the second entity. For example, the new obligation norm may determine whether the first entity transmitted the entirety of a principal payment of a loan to the second entity. The obligation norm associated with the fourth vertex 1031 may be associated to a satisfaction norm represented by a fifth vertex 1041 or a failure norm represented by a sixth vertex 1042.
In some embodiments, advancement of the state represented by the directed graph 1010 to the state represented by the directed graph 1020 or the state represented by the directed graph 1030 may be simulated using a symbolic AI system. For example, the state represented by the directed graph 1010 may be copied into a symbolic AI model, where both the conditional statements associated with the nodes and of the directed graph the edges connecting the nodes of the directed graph may be copied. A symbolic AI system may then simulate state changes using the symbolic AI model to determine an expected value for a smart contract that has already reached the state represented by the directed graph 1010, where the expected value may be a multi-iteration score.
In some embodiments, each of the smart contracts represented by the directed graphs 610, 650, 710, and 1010 may be analyzed using a symbolic AI system to determine one or more multi-protocol scores. For example, each of the smart contracts represented by the directed graphs 610, 650, 710, and 1010 may be analyzed to produce multi-iteration scores such as average scores for each smart contract and a kurtosis value of expected scores. In some embodiments, the analysis may use the same rules to govern the behavior entities in the smart contract by basing the rules on logic types and vertex statuses instead of the contexts of specific agreements. For example, each smart contract simulation may be simulated with a set of rules that include a rule that the probability that a rights norm to cure is triggered instead of a rights norm to accelerate being triggered is equal to 90%. The multi-iteration scores may then be further analyzed to determine a multi-protocol score. For example, based on a multi-iteration score representing a risk score associated with each of the smart contracts, the total exposed risk of a first entity with respect to a second entity may be determined, where the total exposed risk may be a multi-protocol score.
In some embodiments, a directed acyclic graph of cryptographic hash pointers may be used to represent the tamper-evident data store 1102. Some or all of the nodes of the directed acyclic graph may be used to form a skip list or linked list, such as the node corresponding to or otherwise representing as block 1104. In some embodiments, each block represented by a node of this list may include multiple values as content. For example, each respective block may include a timestamp of creation 1106, a cryptographic hash of content of the previous node pointed to by an edge connecting those nodes 1108, a state root value 1110 for a trie of cryptographic hash values that may be referred to as a state trie 1118, a cryptographic hash 1112 that is a root value of a receipt trie 1124 of cryptographic hash values referred to as a receipt trie, and a cryptographic hash value 1114 that is a root value of a trie of cryptographic hash values referred to as a transaction trie 1122. In some embodiments, the block 1104 may be connected to a plurality of tries (e.g., three or more tries) via cryptographic hash pointers. For example, the block 1104 may be connected to Merkle roots (or other roots) of the plurality of tries of cryptographic hash values.
In some embodiments, the state trie 1118 may include multiple levels of cryptographic hash pointers that expand from a root to leaf nodes through 2 or more (e.g. 3, 11, 5, 6, etc.) hierarchical levels of branching. In some embodiments, an account address of a smart contract or instance of invocation thereof may correspond to a leaf nodes, where the smart contract may be an instance of the smart contract described in one or more operations of one or more processes described in this disclosure. In some embodiments, leaf nodes or paths to the leaf nodes of the state trie 1118 may include the fields in the account object 1126. The address may be a smart contract address or instance of invocation of the smart contract, the nonce value may be a count of the times that the smart contract was invoked, the code hash value may be or otherwise include a cryptographic hash of a bytecode representation of the smart contract 1130, the storage hash may be a root (e.g. Merkle root) of a trie of cryptographic hash pointers 1120. In some embodiments, the trie of cryptographic hash pointers 1120 may store key-value pairs encoding a transient program state of the smart contract that changes or is not needed between invocations of the smart contract. In some embodiments, the fields of the account object 1126 may include a predecessor pointer that points to a previous entry of an earlier state trie corresponding to a previous invocation of the smart contract and associated information or hashes.
In some embodiments, the decentralized computing platform 1200 may be private, which may allow a peer computing node of the decentralized computing platform 1200 to authenticate itself to the other computing nodes of the decentralized computing platform 1200 by sending a value based on a private cryptographic key, where the private cryptographic key may be associated with a permissioned tenant of the decentralized computing platform 1200. While
Computer system 1300 may include one or more processors (e.g., processors 1310a-1310n) coupled to System memory 1320, an input/output I/O device interface 1330, and a network interface 1340 via an input/output (I/O) interface 1350. A processor may include a single processor or a plurality of processors (e.g., distributed processors). A processor may be any suitable processor capable of executing or otherwise performing instructions. A processor may include a central processing unit (CPU) that carries out program instructions to perform the arithmetical, logical, and input/output operations of computer system 1300. A processor may execute code (e.g., processor firmware, a protocol stack, a database management system, an operating system, or a combination thereof) that creates an execution environment for program instructions. A processor may include a programmable processor. A processor may include general or special purpose microprocessors. A processor may include one or more microcontrollers. A processor may receive instructions and data from a memory (e.g., System memory 1320). Computer system 1300 may be a uni-processor system including one processor (e.g., processor 1310a), or a multi-processor system including any number of suitable processors (e.g., 1310a-1310n). Multiple processors may be employed to provide for parallel or sequential execution of one or more portions of the techniques described herein. Processes, such as logic flows, described herein may be performed by one or more programmable processors executing one or more computer programs to perform functions by operating on input data and generating corresponding output. Processes described herein may be performed by, and apparatus can also be implemented as, special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit). Computer system 1300 may include a plurality of computing devices (e.g., distributed computer systems) to implement various processing functions.
I/O device interface 1330 may provide an interface for connection of one or more I/O devices 1360 to computer system 1300. I/O devices may include devices that receive input (e.g., from a user) or output information (e.g., to a user). I/O devices 1360 may include, for example, graphical user interface presented on displays (e.g., a cathode ray tube (CRT) or liquid crystal display (LCD) monitor), pointing devices (e.g., a computer mouse or trackball), keyboards, keypads, touchpads, scanning devices, voice recognition devices, gesture recognition devices, printers, audio speakers, microphones, cameras, or the like. I/O devices 1360 may be connected to computer system 1300 through a wired or wireless connection. I/O devices 1360 may be connected to computer system 1300 from a remote location. I/O devices 1360 located on remote computer system, for example, may be connected to computer system 1300 via a network and network interface 1340.
Network interface 1340 may include a network adapter that provides for connection of computer system 1300 to a network. Network interface 1340 may facilitate data exchange between computer system 1300 and other devices connected to the network. Network interface 1340 may support wired or wireless communication. The network may include an electronic communication network, such as the Internet, a local area network (LAN), a wide area network (WAN), a cellular communications network, or the like.
System memory 1320 may be configured to store program instructions 1324 or data 1315. Program instructions 1324 may be executable by a processor (e.g., one or more of processors 1310a-1310n) to implement one or more embodiments of the present techniques. Program instructions 1324 may include modules of computer program instructions for implementing one or more techniques described herein with regard to various processing modules. Program instructions may include a computer program (which in certain forms is known as a program, software, software application, script, or code). A computer program may be written in a programming language, including compiled or interpreted languages, or declarative or procedural languages. A computer program may include a unit suitable for use in a computing environment, including as a stand-alone program, a module, a component, or a subroutine. A computer program may or may not correspond to a file in a file system. A program may be stored in a portion of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub programs, or portions of code). A computer program may be deployed to be executed on one or more computer processors located locally at one site or distributed across multiple remote sites and interconnected by a communication network.
System memory 1320 may include a tangible program carrier having program instructions stored thereon. A tangible program carrier may include a non-transitory, computer-readable storage medium. A non-transitory, computer-readable storage medium may include a machine readable storage device, a machine readable storage substrate, a memory device, or any combination thereof. Non-transitory, computer-readable storage medium may include non-volatile memory (e.g., flash memory, ROM, PROM, EPROM, EEPROM memory), volatile memory (e.g., random access memory (RAM), static random access memory (SRAM), synchronous dynamic RAM (SDRAM)), bulk storage memory (e.g., CD-ROM and/or DVD-ROM, hard-drives), or the like. System memory 1320 may include a non-transitory, computer-readable storage medium that may have program instructions stored thereon that are executable by a computer processor (e.g., one or more of processors 1310a-1310n) to cause the subject matter and the functional operations described herein. A memory (e.g., System memory 1320) may include a single memory device and/or a plurality of memory devices (e.g., distributed memory devices). Instructions or other program code to provide the functionality described herein may be stored on a tangible, non-transitory, computer-readable media. In some cases, the entire set of instructions may be stored concurrently on the media, or in some cases, different parts of the instructions may be stored on the same media at different times.
I/O interface 1350 may be configured to coordinate I/O traffic between processors 1310a-1310n, System memory 1320, network interface 1340, I/O devices 1360, and/or other peripheral devices. I/O interface 1350 may perform protocol, timing, or other data transformations to convert data signals from one component (e.g., System memory 1320) into a format suitable for use by another component (e.g., processors 1310a-1310n). I/O interface 1350 may include support for devices attached through various types of peripheral buses, such as a variant of the Peripheral Component Interconnect (PCI) bus standard or the Universal Serial Bus (USB) standard.
Embodiments of the techniques described herein may be implemented using a single instance of computer system 1300 or multiple computer systems 1300 configured to host different portions or instances of embodiments. Multiple computer systems 1300 may provide for parallel or sequential processing/execution of one or more portions of the techniques described herein.
Those skilled in the art will appreciate that computer system 1300 is merely illustrative and is not intended to limit the scope of the techniques described herein. Computer system 1300 may include any combination of devices or software that may perform or otherwise provide for the performance of the techniques described herein. For example, computer system 1300 may include or be a combination of a cloud-computing system, a data center, a server rack, a server, a virtual server, a desktop computer, a laptop computer, a tablet computer, a server device, a client device, a mobile telephone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a vehicle-mounted computer, or a GPS device, or the like. Computer system 1300 may also be connected to other devices that are not illustrated, or may operate as a stand-alone system. In addition, the functionality provided by the illustrated components may in some embodiments be combined in fewer components or distributed in additional components. Similarly, in some embodiments, the functionality of some of the illustrated components may not be provided or other additional functionality may be available.
Those skilled in the art will also appreciate that while various items are illustrated as being stored in memory or on storage while being used, these items or portions of them may be transferred between memory and other storage devices for purposes of memory management and data integrity. Alternatively, in other embodiments some or all of the software components may execute in memory on another device and communicate with the illustrated computer system via inter-computer communication. Some or all of the system components or data structures may also be stored (e.g., as instructions or structured data) on a computer-accessible medium or a portable article to be read by an appropriate drive, various examples of which are described in this disclosure. In some embodiments, instructions stored on a computer-accessible medium separate from computer system 1300 may be transmitted to computer system 1300 via transmission media or signals such as electrical, electromagnetic, or digital signals, conveyed via a communication medium such as a network or a wireless link. Various embodiments may further include receiving, sending, or storing instructions or data implemented in accordance with the foregoing description upon a computer-accessible medium. Accordingly, the present techniques may be practiced with other computer system configurations.
As described above, some embodiments may predict an outcome score based on program state data. Some embodiments may provide mechanisms to determine counterparty entity actions based on the outcome score and determine possible higher-reward actions for an entity by modeling graph evolution over time. Some embodiments may perform operations, such as those described further below, to predict graph evolution over time based on the state of a smart contract program using one or more machine learning operations.
Graph Evolution and Outcome Determination for Graph-Defined Program States
In some embodiments, a program state of a smart contract program encoded in a symbolic AI system may include a directed graph, where entities of the symbolic AI system may cause events that simulate evolving program state by selecting vertices of the directed graph. Event-causing operations performed by an entity to establish, maintain, or end relationships between entities may be analyzed by an intelligent agent being executed by a computing system. An intelligent agent may include any set of functions, operations, routines, or applications that may perceive or otherwise obtain input values, refer to one or more stored parameters, and determine an outcome response from the input values using the one or more stored parameters, wherein at least one routine of the intelligent agent allows for one or more of the set of stored parameters to be changed over time. The intelligent agent may take advantage of the directed graph and its associated categorizations to determine a set of outcome program states after an event caused by a first entity or a counterparty entity, where the events caused by the counterparty entity are not in the control of the first entity. Using the set of outcome program states, the intelligent agent may respond to counterparty-caused events.
Operations to determine outcome states of a multiparty smart contract program are often prevented by the difficulty of accounting for information asymmetry, differing goals amongst entities, and computational complexity. In some embodiments, an entity of a multiparty smart contract program may perform an action without having the information necessary for precisely valuing the action. In some embodiments, an entity of a multiparty smart contract program may have different goals or strategies with respect to another entity of the program. Furthermore, an entity or its counterparty entities of a multiparty smart contract program may have a large number of options and variations of those options available when causing a state-changing event, which may increase the difficulty of decision-making operations for AI systems tasked with predicting the outcomes of those events. By using a symbolic AI model that models entity transactions using a directed graph with associated vertex categories, a system may more efficiently and accurately overcome these difficulties to determine outcome states and their associated outcome stores.
A smart contract program implemented using a symbolic AI model may include vertex categories associated with norms. The use of vertex categories in association with the norms or norm vertices of a symbolic AI model may increase the speed and accuracy of outcome determination. The categorization of the norms or their associated directed graph vertices into a set of vertex categories such as “obligations,” “rights,” “prohibitions,” or the like may indicate outcome behaviors in response to their conditional statements (“conditions”) being satisfied or not satisfied. By pre-categorizing a set of vertices of a directed graph representing the program state of a smart contract program into one of a set of vertex categories, the system may rapidly simulate evolving a program state into possible future states. In addition, the system may be free from computing burdens related to examining each individual condition and conditional outcome of every norm encoded in a smart contract program or parsing ad-hoc coded statements. Such advantages may be especially useful in many scenarios involving a graph of a symbolic AI model that includes a large number of vertices, such as more than 100 vertices or more than 1000 vertices.
In some embodiments, additional operations may be performed to determine outcome scores, determine counterparty actions, update a directed graph, or retrieve data from a directed graph. Some embodiments may perform such operations or other operations using methods or systems described in the co-pending PCT application bearing attorney docket number “053173-0515078” titled “GRAPH-MANIPULATION BASED DOMAIN-SPECIFIC EXECUTION ENVIRONMENT,” PCT application bearing attorney docket number “053173-0515079” titled “GRAPH OUTCOME DETERMINATION IN DOMAIN-SPECIFIC EXECUTION ENVIRONMENT,” PCT application bearing attorney docket number “053173-0515080” titled “MODIFICATION OF IN-EXECUTION SMART CONTRACT PROGRAMS,” and PCT application bearing attorney docket number 053173-0515081 titled “GRAPH EVOLUTION AND OUTCOME DETERMINATION FOR GRAPH-DEFINED PROGRAM STATES,” which were filed on 2020 Sep. 8 and assigned to the applicant, “Digital Asset Capital, Inc.,” and which are herein incorporated by reference. Some embodiments may further perform operations such as scoring entities, using hybrid systems to efficiently query data, determine outcome data based on an event with respect to multiple directed graphs. Some embodiments may perform such operations or other operations using methods or systems described in the co-pending US patent application bearing attorney docket number “053173-0515218” titled “EVENT-BASED ENTITY SCORING IN DISTRIBUTED SYSTEMS,” US patent application bearing attorney docket number “053173-0515223” titled “CONFIDENTIAL GOVERNANCE VERIFICATION FOR GRAPH-BASED SYSTEM,” US patent application bearing attorney docket number “053173-0515224” titled “HYBRID DECENTRALIZED COMPUTING ENVIRONMENT FOR GRAPH-BASED EXECUTION ENVIRONMENT,” and US patent application bearing attorney docket number 053173-0515226 titled “MULTIGRAPH VERIFICATION,” which were filed on 2020 Sep. 8, and are assigned to the applicant, “Digital Asset Capital, Inc.,” and which are herein incorporated by reference. Some embodiments may perform operations such as dimensionally reducing graph data, querying a data structure to obtain data associated with a directed graph, perform transfer learning operations, or efficiently notify entities. Some embodiments may perform such operations or other operations using methods or systems described in the co-pending US patent application bearing attorney docket number “053173-0508429” titled “GRAPH-BASED PROGRAM STATE NOTIFICATION,” US patent application bearing attorney docket number “053173-0508434” titled “DIMENSIONAL REDUCTION OF CATEGORIZED DIRECTED GRAPHS,” and US patent application bearing attorney docket number “053173-0508433” titled “QUERYING GRAPH-BASED MODELS,” which were filed on 2020 Sep. 8, and are assigned to the applicant, “Digital Asset Capital, Inc.,” and which are herein incorporated by reference.
In some embodiments, one or more operations of the process 1400 may be otherwise initiated in response to a detected change in a program state of an application. For example, an event obtained by a symbolic AI system may indicate that an asset has been removed from a set of assets, which results in the change of an asset value stored in the program state data of the symbolic AI system. In some embodiments, as further described below, in addition to initiating one or more operations of the process 1400, one or more parameters of an intelligent agent associated with an entity may be updated based on the changed program state. Alternatively, or in addition, operations of the process 1400 may begin without detecting a change in the program state of the application.
In some embodiments, the process 1400 includes obtaining a current program state of a symbolic AI model associated with a plurality of entities, as indicated by block 1404. In some embodiments, the symbolic AI model may include a smart contract program, such as a smart contract described above for the operations of block 304. The program state of the symbolic AI model may include a directed graph, such as the directed graphs described above. The use of “norm” in the context the process 1400 may include or be otherwise associated with a norm vertex of a directed graph representing some or all of a program state of the symbolic AI model. In some embodiments, the current program state may indicate an uninitialized form or otherwise un-triggered form and may include a directed graph including a single active vertex without a previously-triggered vertex. Alternatively, the current program state may have already progressed from an initial and may include a set of vertices that have already been triggered. In some embodiments, the directed graph of a current program state may have a few number of vertices, such as one vertex, two vertices, less than five vertices, less than ten vertices, or the like. In some embodiments, the directed graph of a current program state may have a large number of vertices, such as 100 vertices, 1000 vertices, 10,000 vertices, or the like.
In some embodiments, the process 1400 includes determining one or more program states and parameters for using, training, or modifying an intelligent agent based on the current program state, as indicated by block 1408. As further described below, an intelligent agent may include a set of programs, routines, sub-routines, or the like being executed by the system to determine outcome program states or their associated outcome scores based on an input program state. In some embodiments, a program state used for or by an intelligent agent may be identical to the current program state. Alternatively, a program state used in the process 1400 may be modified from a current program state, where some values of the modified program state may differ from the values of the current program state. In some embodiments, the program state in the process 1400 may include changes in values to simulate additional possibilities at a current time or in the future. For example, a first entity may be engaged with the second entity in a smart contract program to stream a computing resource to the second entity. Even if the current program state indicates that the computing resource is available, the program state used in the process 1400 may include a program state representing a scenario in which the computing resource becomes corrupted while being used by the second entity. The smart contract program may include a rights norm that allows the second entity to provision additional computing resources in such an event, which may allow this scenario to be tested. As described further below, multiple scenarios may be used to determine action values across multiple possible scenarios. In some embodiments, probabilities or other measures of likelihood may be assigned to scenarios. A richer set of possible outcome program states determined from these associated measures of likelihood may be used to account for possible future outcomes.
As described below, a scenario may be determined based on actions performed by additional entities. For example, if a first smart contract program includes a set of obligation norms and rights norms associated with a first entity and a second entity, the actions of a third entity that is not an entity listed in the first smart contract program may still affect the program state of the first smart contract program. These effects may manifest by altering a parameter of the program state, such as increasing an amount of required material density in a manufacturing context or alter a specific state of an entity. For example, a failure of the third entity to transfer an asset or score value to the second entity may likewise affect the program state of the first entity by altering one or more expected reward values associated with the values being transferred by the second entity. As another example, if the second entity is noted to fail an obligation to the third entity, a parameter associated with the second entity may be modified to indicate that the second entity has a higher likelihood of failing an obligation to the first entity.
In some embodiments, the process 1400 may include categorizing a program state value using one or more categorization operations, as indicated by block 1412. The program state value may be a numerical value, a categorical value, a set of values, or the like. The categorized program state may include one or more converted values that are based on the one or more program state value. For example, a first score value in the program state that is equal to “30” may be converted to a categorical value using a categorization operation that bins all quantitative values between 0 to 15 into the category “LOW” and all quantitative values greater than 15 into the category value “HIGH,” resulting in the conversion of the program state value from 30 into the categorized program state value “HIGH.” Using the one or more categorization operations may result in a discretized representation of a program state value, which may reduce the computation resources required to generate counterparty predictions. Furthermore, an intelligent agent may account for categorized program states during training or prediction operations either by applying an initial simulated event to a program state and then modifying the program state using a categorization operation or by applying the categorization operation to the simulated event directly.
In some embodiments, the process 1400 may include determining a set of possible state-changing events causable by an entity based on the program state, as indicated by block 1416. In some embodiments, the set of possible state-changing events causable by an entity may be determined based on conditions associated with a set of active vertices associated with norms. Determining the possible state-changing events causable by the entity may include finding the set of vertices indicated as active in the program state (“set of active vertices”). For example, the program state may include a set of active vertices, where the set of active vertices are associated with a set of norms that include an obligation norm to allocate memory, a rights norm to inspect an asset provided in response to the allocated memory, and a prohibition norm on deleting allocated memory. The term “active vertex” means that the norm associated with the active vertex is triggerable such that satisfying a set of conditional statements (“set of conditions”) associated with the norm will result in the system performing one or more operations conditional on the set of conditions being satisfied.
In some embodiments, the set of possible state-changing events causable by an entity may also include state-changing events associated with norms that are not active in the current program state. For example, a first norm may be a rights norm to terminate a smart contract that may be activated if a specified type of encrypted message is sent by a third entity via a specific API. A current program state may be such that the specified type of encrypted message has not been sent, but an intelligent agent may use a scenario that includes a probability that the specified type of encrypted message will be sent in the future. For example, the scenario may include that the probability of the specified type of encrypted message will be sent within the next two months is equal to 0.15%. As further described below, during traversal of a directed graph of the program state that spans one or more edges, an intelligent agent may simulate the encrypted message being sent and include its subsequent outcome states as part of an expanded directed graph. The outcome states and their associated outcome scores may then be included and considered by the intelligent when providing projections of outcomes.
In some embodiments, a plurality of events may trigger one or more of a set of active vertices associated with active norms. For example, a first state-changing event may include allocating 10 gigabytes (GB) from the first entity to a second entity and a second state-changing event may include allocating 50 GB from the first entity to the second entity, where either the first state-changing event or second state-changing event may both satisfy a same obligation norm. In some embodiments, the system may simulate and keep both events as separate events and, as further described below, determine different action values for each of the selected set of events, where each of the selected set of events may be associated with an action of an entity. Alternatively, some embodiments may simplify multiple possible events that satisfy a norm to one or more categorized events, such as providing a minimum or maximum amount of allocated memory necessary to satisfy a norm and determine an action score for the categorized event. In some embodiments, as further described below, simulations of future program states may directly trigger norms of a program state, where the conditions associated with triggering the norms may be retroactively applied to determine the event that triggered the norm. In some embodiments, the selected set of events may be determined based on probability distributions associated with the entities. For example, a first entity may include a set of entity properties defining a gaussian distribution, where a mean memory allocation defined by the gaussian distribution may be used to determine an amount of memory allocated in an event of the selected set of events.
While the above example describes an event to allocate memory, other events are possible. For example, some embodiments may include events such as an allocation of a transfer rate in a data pipeline, a transmission of instructions to accelerate a payment deadline via an API, or transference of an asset from a first entity to a second entity. For example, in response to the determination that an active vertex is associated with an obligation norm for a first entity to pay a minimum amount, the set of possible events may include a first possible event caused by the first entity transferring the minimum amount, a second possible event caused by the first entity to transferring 200% of the minimum amount, and a third possible event caused by the first event transferring a remaining balance in full. In some embodiments, as further described below, a learning system may be used to simulate the occurrence of multiple events that lead to a same directed graph shape but may result in different final program states.
In some embodiments, the process 1400 includes updating a set of action values or other parameters of an intelligent agent for the set of possible state-changing events based on parameters of an intelligent agent or a set of vertex categories of the symbolic AI model, as indicated by block 1424. In some embodiments, the set of parameters of an intelligent agent may include parameters defining a policy, where a policy may include a policy function that maps a program state to an event caused by an entity or action or a probability of the event occurring. In some embodiments, the policy function may be deterministic and predict the same action when presented with the same program state. Alternatively, the policy function may be stochastic, wherein a same program state may result in different predicted actions based on a probability distribution. For example, when provided a first state as an input, the policy function may provide 25% as the possibility for the first state to transition to a second state and 75% as the probability for the first state to transition to a third state.
In some embodiments, an action value may include a quantitative or categorical value. An action value may be associated with an event caused by an entity and, as further described below, and may be used to determine outcome states or response instructions. Alternatively, an action value for an entity may be associated with an action of the entity, where the action may be analyzed to determine which elements of the action change a program state. For example, a first entity may perform an action that includes transferring an amount to the second entity along with a message indicating that no further transfers will occur. In some embodiments, a system may ignore the message as having no effect on a program state and consider the event associated with the action as identical to the first entity transferring the amount to the second entity without sending an associated message. In some embodiments, updating an action value may include changing an existing value associated with the action value. Alternatively, updating the action value may include determining a new action value and associating it with the possible state-changing events causable by an entity. In some embodiments, the action value may be modified or otherwise determined based on an entity property, as further described in this disclosure. For example, an entity property may modify the likelihood of events occurring, modify a reward value associated with a possible outcome of an action, or otherwise determine a behavior pattern associated with an entity.
In some embodiments, the parameters of the intelligent agent may include the action values themselves. For example, the intelligent agent may use the action values as parameters of the policy function by including an operation to select which event is most likely to be caused by an entity based on the action values. The policy function may select the event from a set of possible events by determining which event has the greatest action value from the action values associated with the set of possible actions.
In some embodiments, the action value associated with a set of possible state-changing events causable by an entity may be determined based on a reward value associated with a change from a first program state to a second program state caused by the state-changing event. The reward value may be equal to or otherwise based on a score change of an entity and may be associated with one or more conditions of the set of conditions. For example, if a condition to satisfy an obligation norm includes a threshold value of 10 units, where the threshold value indicates that satisfying the obligation norm requires an allocation of 10 computing instances to a first entity from a second entity, the reward value for the obligation norm or its associated directed graph vertex may be +10 for the second entity. In some embodiments, each entity may have its own set of reward values and, correspondingly, their own set action values based on those reward values. For example, a first entity may have +10 as the reward value associated with a first vertex based on a threshold value of 10 for a condition associated with the first vertex, and a second entity may have −20 as the reward value associated with the same first vertex based on the threshold value and an addition −10 as part of a transaction cost associated with the first vertex. In some embodiments, the set of vertex categories may be used to modify or set reward values based on their failure or satisfaction. For example, the system may determine that a first vertex is categorized as an obligation and, in response to a status change indicating that the first vertex is failed, deduct a failure penalty from the reward value associated with the first vertex or a child vertex of the first vertex.
In some embodiments, a reward value may be modified based on an entity property value associated with an entity or entity role. The entity property value may be used to change the reward value to a greater or lower value. An entity property may indicate an entity's characteristic, limitation, behavior pattern, or the like. For example, a first entity may be associated with a first entity property for an entity property correlated with an acceptable loss before catastrophic entity failure (“loss multiplier threshold”), where losses greater than the first entity property value are multiplied by five (or some other numeric value). For example, if the first entity property value for a loss multiplier threshold is 200, losses greater than 200 will be multiplied by five, such that the reward value for a loss of “200” may be equal to “−200” for the first entity, but a reward value for a loss of “201” may be equal to “1005.” Alternatively, or in addition, instead of applying a multiplier to a value, some embodiments may set a reward value to a pre-determined value. As further discussed below, by altering reward values based on one or more entity properties, entity biases, tendencies, or behavior patterns may be simulated. Alternatively, or in addition, a probability distribution defined by one or more entity properties may be used to modify or otherwise determine a reward value.
By using a threshold value encoded in or otherwise associated with the conditions associated with norm vertices, efficient estimation of reward values may be performed without requiring additional and potentially error-prone simulations of unexpected events or entity behaviors which are unlikely to occur or whose variations are otherwise not relevant to an outcome program state or outcome score. In some embodiments, the threshold value of a condition itself may be stored in a conditions data model and may be retrieved based on an identifier of a vertex or norm associated with the condition. By storing the threshold value in retrievable form, operations to use the threshold values of conditions in a symbolic AI model may be performed with greater speed and accuracy. In addition, other values associated with vertices may be stored and retrieved based on vertex or norm identifiers, such as transaction costs associated with transactions between entities.
In some embodiments, the action value may be determined based on future reward values associated with possible future program state transitions and may further be modified based on one or more discount factors, transition probabilities, environmental state effects, learning rates, or some function thereof. For example, an action value may be an output of a policy function. The action value may be affected by various other program state values, intelligent agent parameters, or other values. The future reward values may be based on a set of threshold values encoded in or otherwise associated with the norms simulated to as triggered in the future program state. In addition, the reward values may be based on additional factors, such as other program state values, specific directed graph configurations, types of vertex statuses, or the like. In some embodiments, determining a set of action values may include generating a sequence of program states that include an initial program state, a set of intermediate program states, and a terminal program state that indicates that a smart contract program is complete. For example, an action value A may be determined using statement 12 below during the training of an intelligent agent. As shown below, A(st, et) is the action value associated with the event et being caused while the program state is in the state st, α is a learning rate, rt+1 is the reward value associated with the state change caused by event et from the program state st, and γ is a discount factor:
A(st,et)←A(st,et)+α[rt+1+γA(st+1,et+1)−A(st,et)] (12)
In some embodiments, operations to determine an action value for a state-changing event or action may include expanding a directed graph of a first program state to include its possible child vertices to generate a second directed graph representing a possible future program states. The possible future program states may be reached from a set of possible events and may be used to determine a set of reward values associated with the set of possible events. The system may determine an action value based on the score changes of that state, a discount factor, transition probabilities to possible future states, and the path scores associated with the possible future states. The reward value may be determined in various ways. For example, statement 13 below may be used to determine a path score as a total reward value Rπ(s) associated with a state s, where Rπ(s′) is a total reward value associated with a possible state s′, s′ and e may represent an index of each possible program state or event caused by an entity, respectively, r is a reward value that may indicate a local, immediate reward corresponding to the transition from the state s to the state s′, γ is a discount factor, p is a probability value of transitioning from the state s to the state s′, and π represents a policy function that provides a probability of an entity causing event e when at the program state is at state s:
As shown above for statement 13, some embodiments may determine the total reward value for a current state based on the expected rewards and corresponding probabilities of occurrence of future states. For example, the system may traverse through the directed graph of a smart contract program and determine a first reward value based on a determination that a first program state corresponds with an event caused by the first entity satisfying an obligation norm to transfer 300 units to a second entity as part of a resource allocation agreement. The system may assign a reward value of −300 in association with the first entity and a reward value of +300 in association with the second entity. In some embodiments, these reward value may be used to represent score changes in one of various possible scenarios, such as score changes in an agreement by the first entity to allocate 300 terabytes to the second entity in return for the second entity reallocating more memory to the first entity at a later time. Some embodiments may use the rewards to represent other score changes, such as changes during transfers of digital currency, reputation scores, or the like. If the satisfaction of this obligation does not result in the termination of the smart contract program but instead in the activation of a first rights norm for the second entity to transfer 400 units to the first entity within fifty days and a second rights norm for the second entity to transfer 310 units to the first entity within ten days.
In some embodiments, a time difference may be used in conjunction with the discount factor to determine a future value. The system may use a discount factor of 99.5% per day and determine that, based on this discount factor and reward value, the discounted reward value for the first possible state value “γRπ(s′1)” is 311.33 by computing the result of the expression “0.99550×400.” Similarly, the system may use a discount factor of 99.5% per day and determine that, based on this discount factor and reward value, the discounted reward value for the second possible state value “γRπ(s′2)” is 294.84 by computing the result of the expression “0.99510×310.” In some embodiments, the system may use of a policy function that determines that second entity has a 95% chance of satisfying the first rights norm and a 5% chance of satisfying the second rights norm in statement 13 above to determine a total value for the state Rπ by computing a total reward value of 10.5 by simplifying the expression “−300+95%×311.33+5%×294.84.” The reward value may be further modified based on vertex categories. For example, some embodiments may impose additional reward value deductions for failing an obligation, violating prohibitions, or the like. Furthermore, reward values may be further altered based on one or more entity properties.
In some embodiments, an end state may be used when determining reward values for possible future states. In some embodiments, an end state may be a terminal program state. A terminal program state for an application may be one in which no additional program states are possible after the application has reached the terminal program state. For example, a function may test whether any norms are triggerable in a program state, and, if the function determines that no norms of the program state are triggerable, determine that the program state is in a terminal program state. In some embodiments, the system may determine that a program state is terminal if a terminal norm is reached, where a norm or its associated vertex may be categorized as a terminal norm to indicate that a terminal program state has been reached.
In some embodiments, an end state may be based on an absolute or relative expiration threshold. For example, some embodiments may include a determination that a program state has reached its expiration threshold of 30 years, where each program state has an associated time point based on a timestamp associated with the satisfaction of the norms that would result in that program state. For example, a program state may include an obligation by the first entity to allocate an amount of computing resources to a second entity every month, and determining a set of events that satisfies these obligation norms may include advancing the simulated time by one month for each event that satisfies an obligation norm. If an expiration threshold of the symbolic AI model is one year, the system may determine that an end state has been reached once the one-year expiration threshold has been reached. Thus, an event that satisfies the twelfth obligation may be associated with a non-zero reward value that affects the action value of a currently active norm, and an event that satisfies the thirteenth obligation norm may not affect the action value of the same currently active norm.
In some embodiments, linear algebra direct solution methods may be impractical for determining reward values due the number of possible states. In addition, assigning value in real-world scenarios with incomplete information may provide additional challenges. Such challenges may include an inability to fully quantify the risks associated with certain types of anticipated rewards, the activation of norms that allow for new possible actions based on changes in an environment, or the like. For example, in response to an electrical blackout that had previously not been occurring, a first-party entity representing a manufacturing center may trigger a rights norm that requires a second entity representing a utility company to allocate a specified amount of electrical energy and terminate the program state, but the satisfaction of such an obligation may be physically impossible. Assigning an action value to a program state may include a determination of the value based on both the specified amount of electrical energy and a probability value associated with non-delivery of the specified amount of electrical energy.
In some embodiments, action values may be determined by performing repeated simulations of applying different sequences of events to a smart contract program and updating action values based on intermediate or final results of the repeated simulations. For example, a system may determine the action value by using dynamic programming, Monte-Carlo simulations, or temporal-difference learning. In some embodiments, an intelligent agent may perform these repeated simulations using one or more machine learning models to determine action values associated with the possible events or actions associated with a program state. Embodiments of these learning operations usable to determine action value(s) and outcome states are described further below for
In some embodiments, the process 1400 includes determining if there are additional scenarios for consideration, as indicated by block 1440. In some embodiments, the system may be configured to consider only a current program state. Alternatively, the system may be configured to consider multiple program states based on the current program state. For example, the system may have a pre-arranged set of scenarios where assets are modified to have greater quantitative resources or fewer quantitative resources. In addition, as discussed further below, the system may use scenarios determined based on environmental variable changes or possible actions from additional entities that are not listed in the entity list of a smart contract program state. For example, the system may update action values based determines that there are additional scenarios for consideration, operations of the process 1400 may proceed to block 1408. Otherwise, operations of the process 1400 may proceed to block 1450.
In some embodiments, the process 1400 includes determining a set of outcome program states or a set of outcome scores using the intelligent agent, as indicated by block 1450. The set of outcome program states may include using the operations above to predict one or more possible outcome program states. For example, an intelligent agent may use the operations above to determine a set of possible directed graphs representing in a set of terminal program states. Alternatively, or in addition, the intelligent agent may predict outcome program states that are not terminal program states.
The set of outcome scores may be associated with the set of outcome program states and may include various types of values. In some embodiments, the set of outcome scores may include a total score change determined from the reward values and the transition probabilities from the current program state to a set of terminal program states. In some embodiments, the set of outcome scores may be based on a starting value associated with an entity and may also be separated based on vertex categories or events. For example, a set of outcome scores for an entity may include a total score owed to the entity based on a first set of obligation norms, a total score owed by the entity based on a second set of obligation norms, a total score provided to the entity via a first set of events, and a total score provided by the entity via a second set of events. Alternatively, or in addition, the set of outcome scores may include a probability weight for a cumulative reward score reaching a target value or for a particular outcome program state or outcome program state value occurring. For example, an outcome score may include a probability weight correlated with the probability that a first entity fails to satisfy an obligation norm, such as by allowing a time point to satisfy a failure time threshold of the obligation norm.
In some embodiments, the system may determine whether an outcome score satisfies a set of outcome score thresholds and, in response, perform an action in response to the outcome score satisfying the set of outcome score thresholds. For example, the system may determine that an outcome score equal to a probability that a set of outcome program states will occur is greater than an outcome score threshold and, in response, transmit a warning message indicating that the outcome score is greater than the outcome score threshold via an API to a device. Alternatively, or in addition, the system may be configured to perform one or more additional actions on behalf of one or more entities in response to a threshold be satisfied. For example, in response to an outcome score satisfying a net loss threshold, the system may be configured to trigger a rights norm that obligates a first entity to return a score value to a second entity. Furthermore, similar to other methods described in this disclosure a plurality of outcome states or outcome scores may be used to determine a population of outcome states, a population of outcome scores, or their corresponding population scores. A population score may include population statistics such as mean values, median values, kurtosis values, tail risk tolerance, or the like.
Alternatively, instead of expanding a directed graph of an application state directly, the system may simulate events that trigger the vertices of the directed graph. As described above, these simulated events may include simulations of one or more entities that cause events based on the norms of the directed graph. For example, the system may simulate the triggering of a first obligation norm associated with a condition requiring that a first entity release at least 500 GB of memory in response to a simulated event of the first entity releasing 850 GB of memory.
In some embodiments, the system may traverse the directed graph to determine a set of score-based reward values, as indicated by block 1508. In some embodiments, the system may traverse the directed graph while expanding the directed graph. For example, as the system expands a directed graph, the system may also record the score changes associated with outcomes of satisfying or failing the norms of the directed graph. An intelligent agent may use these score changes to determine score-based reward values associated with transitions between different program states. For example, failing a vertex categorized as an obligation norm may set a vertex status to indicate that the obligation norm is failed and activate a child vertex associated with a norm condition having a threshold value equal to “−10.” In response, the score-based reward value associated with the state change may be a decrease of 10.
In some embodiments, the system may traverse the directed graph after performing one or more expansion operation on the directed graph in order to determine score-based reward values based on the outcomes of satisfying or failing each norm of the directed graph. As described above, an obligation norm may be failed if it is not satisfied if a time point reaches a time threshold of the obligation norm. Furthermore, during simulation of events, an event may include an event that would satisfy the conditions of a norm that occurs after a failure time threshold, and the intelligent agent may determine that a vertex should be associated with a status indicating failure if response to a determination that the time point associated with the condition is associated with a failure time threshold.
In some embodiments, the system may update a set of data structure instances (e.g., an array, a vector, a list, or the like) to store the set of reward values associated with each program state or program state transition. In addition, the set of data structures may include labels for the program states of an application, the transitions between each of the program states, action values associated with each transition, or the like.
In some embodiments, the process 1500 may include determining a set of training values for an intelligent agent, as described in block 1512. The set of training values may include training inputs and training objectives, where the training inputs and training objectives may be obtained using one or more various methods. An intelligent agent may include one or more machine learning models. Example machine learning models may include supervised learning models, unsupervised learning models, semi-supervised learning models, reinforcement learning models, or the like. In some embodiments, using the machine learning model may include training or using a set of neural networks, such as one or more of the neural networks described above. Various types of neural network training methods may be utilized to determine the parameters of a neural network. In some embodiments, the neural network training method may include a backpropagation operation that includes passing a training input through the neural network to determine a neural network prediction. The neural network prediction may be compared to a training objective to determine a loss function. The loss function may then be propagated back to update the parameters of the neural network using a gradient descent method. This process may be repeated until the neural network prediction sufficiently matches the training objective to update the parameters of the neural network.
In some embodiments, the training inputs and training objectives may be acquired from histories of similar smart contract programs. Alternatively, or in addition, the training inputs and training objectives may be acquired from simulations of possible future events. Furthermore, the training inputs and training objectives may depend in part on the set of learning systems used to determine outcome program states or outcome scores. In some embodiments, training inputs may be iteratively generated during each iteration of training. For example, the intelligent agent may generate a first set of training inputs during a training iteration, where a training iteration may include simulating evolving program state using a pseudorandom method to determine which set of vertices to include in a path through a directed graph. The simulated state evolution may also include updating a set of action values or other parameters based on the results of the simulated state evolution and generating a second set of training inputs based on the updated parameters. The training inputs may be generated during a simulation of the satisfaction of norms or failure to satisfy norms. In some embodiments, the determination of the training inputs may be based on one or more entity properties. For example, an entity may include a set of values defining a probability distribution, where the probability distribution may be used to determine whether the entity will satisfy or fail an obligation, and the system may sample from the probability distribution to determine one or more values for a training input.
In some embodiments, the process 1500 may include determining a set of action values or other set of parameters of the intelligent agent based on the set of reward values and training values, as indicated by block 1516. An action value may represent a weight assigned to a possible action of an entity that causes a state change in the program state of an application. In some embodiments, the action value may be determined based on a set of paths through a directed graph, where each respective path of the set of paths includes a respective plurality of vertices and respective directed edges connecting each vertex of the respective plurality of vertices with at least one other vertex. In some embodiments, a set of action values may be determined based on determining the sequence of reward values associated with the respective set of vertices for each respective path of a set of paths. The action value may be further determined based on training values obtained from historical data or acquired from simulations.
In some embodiments, the intelligent agent may include a model-based reinforcement learning model as a part of a machine learning model. Using a model-based reinforcement learning model may include using a Markov decision process comprising a state transition model to simulate program state evolution by predicting a next program state and a reward model to determine an expected reward during the transition to the next program state. In some embodiments, the system may continue simulating the evolution of a program state through a sequence of states by traversing through a directed path based on the vertices and edges of a directed graph until a terminal program state is reached. The evolution of the program state may be simulated over a set of simulated state evolutions, where each simulated state evolution in the set of simulated state evolutions may be associated with a path through the directed graph. The associated paths may be used to form a set of paths. As a system traverses a directed path, the system may determine a subset of reward values associated with the directed path and may use this subset of reward values to determine a path score or otherwise use the subset of reward values to determine an action score. In some embodiments, the expected reward value between each state change may be based on the set of score-based reward values determined above, where each path in a set of paths may be associated with a different expected total reward value. In addition to score changes associated with vertices, the system may also modify action values based on the satisfaction or failure of obligation norms. For example, the system may decrease the reward value by a failure penalty associated with the transition from a first state to a second state if the transition was the result of failing an obligation norm. In some embodiments, the failure penalty may be a pre-determined value.
Various reinforcement learning models may be implemented to determine the action values using a neural network of the intelligent agent, such as a policy gradient approach, a Q-learning approach, a value-based learning approach, some combination thereof, or the like. In some embodiments, the system may use a policy gradient approach. For example, using a policy gradient approach may include using a set of policy weights equal to or otherwise based on the action values, wherein each policy weight corresponds with a possible event that an entity may cause at one or more program states represented by a configuration of a directed graph. In some embodiments, the intelligent agent may perform repeated iterations of the simulated state evolution using a set of training inputs to determine the results of an entity acting according to instructions based on the policy function, wherein each intermediate outcome or terminal outcome may be used to increase or decrease a corresponding policy weight. The set of simulated state evolutions may being performed continue until a training objective is achieved, where the training objective may include a maximization of a total reward value, a minimization of a loss function, a minimization of the number of obligation norms having a status as being failed, or the like. Various policies may be used, such as softmax policy, Gaussian policy, or the like. In some embodiments, the number of iterations may be greater than 10 iterations, greater than 1000, greater than 10,000 iterations, greater than 100,000 iterations, greater than 1,000,000 iterations, or the like.
In some embodiments, the system may use a Monte Carlo Tree Search (MCTS) operation to determine the set of action values by sequentially selecting vertices of a directed graph based on the graph edges. The system may use a trained neural network to determine an initial set of actions causable by a first entity and their associated initial set of action values. Each action of the initial set of actions may be associated with an initial action value also determined by the trained neural network. The initial action value may be based on the score-based reward values determined above and may be correlated with a probability of the first entity performing that associated action. In some embodiments, the initial action value may be the output of a softmax function. For example, the initial set of actions determined by a neural network may include a first action associated with a pre-processed weight equal to 3, a second action associated with a pre-processed weight equal to 1, and a third action associated with a pre-processed weight equal to 2. The intelligent agent may apply a softmax function to the pre-processed weights to determine an initial set of action values approximately equal to [0.665, 0.090, 0.245].
In some embodiments, a system may perform a comprehensive set of tree search operations through each vertex of a directed graph until every end state is achieved. For example, if there are five paths from a starting vertex to a set of end states, some embodiments may comprehensively traverse each of the five paths to determine five sets of reward values associated with each of the five paths. In some embodiments, the system may determine each possible path from a starting vertex to each terminal vertex and then determining the associated reward values and probabilities associated with the vertices of each respective path to determine action values associated with entity actions. Alternatively, the system may determine multiple sequences of events caused by one or more entities to determine a set of paths through a directed graph such that at least one of the sequences of events is associated with each of the paths and every possible path from the starting vertices to a terminal vertex or another vertex associated with an end state is one of the set of paths.
In some embodiments, a directed graph may be searched using probabilistic methods. For example, if there are five paths from a starting vertex to a set of end states, some embodiments may traverse four of the five paths and avoid simulating a selection of at least one of the vertices of the directed graph. Various implementations of a probabilistic method may be used. For example, action values of a directed graph may indicate that an obligation norm vertex may a 99.99% chance of being satisfied and a 0.01% chance of being failed for simulation. Over the course of 100 searches through a directed graph to simulate evolving program state from an initial program state that starts at the obligation norm vertex, the system may use a random, pseudorandom, or quasi-random number generation method to determine whether the obligation norm vertex is satisfied or failed. Each search through a directed graph may include determining a set of computed values using the random, pseudorandom, or quasi-random number and selecting the next vertex based on the set of computed values. In some embodiments, each search may continue until a terminal vertex or some other vertex associated with an end state is reached. The search may be repeated for multiple iterations. For example, a search may be repeated for a directed graph with new computed pseudorandom values to determine a corresponding new path through the directed graph more than 10 times, more than 1000 times, more than 10,000 times, more than 100,000 times, more than 1,000,000 times, or the like.
In some embodiments, it is possible that the use of a random, pseudorandom, or quasi-random number generation method to select paths based on these action values may skip one or more vertices such that no paths that include activated vertices resulting from failing the obligation norm. Various random, pseudorandom, or quasi-random number generation methods may be used, such as a method based on a physical phenomenon or a computational method based on a seed value. Example random, pseudorandom, or quasi-random number generation methods may include wideband photonic entropy source-dependent methods, linear congruential generator methods, middle square methods, middle square Weyl sequence methods, permuted congruential generator methods, low-discrepancy sequence methods, or the like.
In some embodiments, the neural network may include a deep neural network usable as a fuzzy store of values. For example, the deep neural network may be a deep convolutional neural network, deep graph neural network, deep recurrent neural network, some combination thereof, or the like. The neural network may receive the directed graph vertices and edges as input and be trained to predict a return value based on the reward values determined above. In some embodiments, the training objective may be based on a total reward value. Alternatively, or in addition, the training objective may be based on the satisfaction of each obligation norm or non-satisfaction of a prohibition norm. For example, a training objective may be classified as positive or negative, where a positive outcome is associated with a terminal program state in which all obligation norms are satisfied, and a negative outcome is associated with all other possible terminal program states. Alternatively, or in addition, a training objective may be based on both reward values and vertex categories. For example, a positive outcome may be associated with all terminal program states in which all obligation norms are satisfied or over 100% of a target score threshold is satisfied, where the target score threshold is based on score transfers determined from the reward values.
After performing each of a set of searches in a set of MCTS operations, the intelligent agent may determine a path score associated with the path taken through the directed graph based on the reward values associated with the vertices of a directed path. A directed path may include a linkage of graph vertices via their connected directed edges, where the path direction may be determined by the edge directions of the directed edges. For example, if performing an iteration of the MCTS operation resulted in a first path that started a first vertex associated with a first reward value equal to “−400,” a second reward value of “−100,” and a third reward value of “+1100,” the system may determine that a total reward value is equal to “+600” by summing the three individual reward values. As further described above, the system may also include vertex category based modifications to the reward value. For example, if the system determines that the path includes a norm vertex having a status indicating failure, the system may reduce the total reward value from “+600” to “−9000” based on a failure penalty equal to “−9600.” The intelligent agent may then update the aggregate score associated with the action based on the total reward value. After a loop through an MCTS operation, the system may also retrain the neural network based on the aggregate score or total reward values to determine the values for a next run-through of the MCTS operation.
In some embodiments, a deep Q-learning approach may be taken to implement a reinforcement learning method. For example, an intelligent agent using a deep Q-learning approach may use a multi-layer neural network to determine internal reward values based on a program state. The intelligent agent may include one or more various types of neural networks, such as a convolutional neural network or a graph neural network. Graph neural networks may include graph auto-encoders, graph convolutional neural networks, graph generative networks, graph attention networks, and graph spatial-temporal networks. In some embodiments, a graph neural network may be useful to receive the structure of a directed graph representing a smart contract program state as an input to generate one or more output values. During training, the intelligent agent may also store previous program states, program state transitions (or the action used to cause the transition), reward values, and outcome program states as experiences in a memory buffer and select these experiences from the memory buffer for retraining the neural network.
In some embodiments, the intelligent agent may implement a counterfactual regret minimization method when determining an action value. Use of a counterfactual regret minimization method may include operations described in “Deep Counterfactual Regret Minimization” (Brown, Noam; Lerer, Adam; Gross, Sam; Sandholm, Tuomas. Nov. 1, 2018, arXiv: 1811.00164), which is hereby incorporated by reference. To implement a counterfactual regret minimization operation, the parameters of the intelligent agent may include one or more strategy profile values and a set of counterfactual regret values, wherein each counterfactual regret value of the set of counterfactual regret values is associated with an action. In some embodiments, the counterfactual regret value may reduce a corresponding action value and may represent the regret of not following a strategy profile. For example, a counterfactual regret value for an action may be quantitatively defined as a difference between a total expected loss of an algorithm using a combination of strategies and a minimum total loss when following a single strategy.
In some embodiments, a counterfactual regret may be determined based on a calculation of counterfactual utility values for each of a set of possible events that an entity may cause, where a set of strategy values may be used to assign probabilities to each event of the set of possible events. The strategy values may be determined using a strategy function, where the strategy function for a given program state may be based on a probability distribution function over a set possible events causable by an entity when the program is at the given program state.
Determining a counterfactual utility value of an event caused by an entity may be based on the set of reward values determined above and a set of probabilities. A probability of the set of probabilities may indicate reaching a terminal program state from the current program state after the entity causing for the entity, where the computation may include different paths from a set of paths, where each path of the set of paths may start at an initial vertex and end at a terminal vertex. In some embodiments, determining a counterfactual utility value of an action for a first entity may include a sum of products, where each of the products is a product of a total reward value associated with a terminal program state and a corresponding probability of reaching the respective terminal program state from the current program state. In some embodiments, an initial program state may be changed to one or more intermediate program states before being changed to a terminal program state. For example, a counterfactual regret value ui for entity i may be determined using statement 14 below, where πσ(se,s′) represents the probability of reaching a terminal program state s′ starting at a program state se, where se may represent the subsequent program state after a state-changing event e is received, W is the set of terminal program states, π−iσ(s) represents a counterfactual reach probability of reaching the state s assuming the strategy a is used, and u(s′) represents a path score as determined by incrementing/decrementing the cumulative reward value when changing from the current state s to the final state s′ based on the reward values determined above:
In some embodiments, the system may determine the probability of reaching a respective terminal program state using a set of strategy values, where the strategy values may be based on an information set representing the amount of information about the smart contract program state available to the first entity. In some embodiments, each entity may have all information available about the smart contract program state, resulting in a perfect information set. Alternatively, one or more of the entities may have incomplete information about the smart contract program state.
In some embodiments, the intelligent agent may execute a regret matching routine to calculate a counterfactual regret value for one or more possible actions. Using the regret matching routine may include determining a reward value for an entity associated with the entity causing an event at a time point, where the reward value may be equal to or otherwise based on the score changes caused by the event. Using the regret matching routine may also include determining a total reward associated with causing the event at the time point. Using the regret matching routine may also include determining an additional reward value associated with causing the event based on a set of strategy values. Using the regret matching routine may also include determining an additional reward value associated with causing the event based on a sum of total reward values associated with using multiple sets of strategy values. Using the regret matching routine may also include updating a set of strategy values at the time point based on a regret value for an action and a total regret value across all actions. For example, the regret matching routine may include updating a strategy value based on a ratio of a regret value acquired of performing a first action with a first information set and a sum of regret values, where each regret value of the sum regret values is based on a different action and the first information set.
In some embodiments, additional strategic behavior may be considered by using neural networks to consider the viability of additional events causable by an entity that would be ignored or otherwise grouped in a category. For example, some embodiments may perform one or more operations similar to those used in the MuZero algorithm, where the MuZero algorithm is described in “Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm” (Schrittwieser, Antonoglou, Hubert, Simonyan, Sifre, Schmitt, Guez, Lockhart, Hassabis, Graepel, Lillicrap, Silver. Submitted 19 Nov. 2019. arXiv: 1911.08265), which is hereby incorporated by reference. Using a MuZero algorithm may include performing a set of self-play operations to determine viable event-causing actions and a set of a training operations based on the results of the self-play operations to determine score-maximizing behavior.
In some embodiments, the intelligent agent may store multiple versions of the neural network in a neural network parameter store and outcome program states in an outcome program state store. In some embodiments, the outcome program states may include results from a set of self-play operations or actual outcome program states from previous executions of a smart contract program. During a self-play operation, the intelligent agent may obtain an initial set of events causable by an entity, such as transferring a set of different score amounts at a set of different time points, changing one or more environmental states, or the like. In some embodiments, the intelligent agent may determine an initial set of events causable by an entity based on the conditions of the smart contract program. For example, if a condition of an obligation norm of the smart contract program includes a score transfer threshold requiring the allocation of at least 10 gigabits per second (GB/s) from an entity with respect to a data pipeline, the intelligent agent may vary the score transfer threshold to a set of modified values including 50% of the score transfer threshold and 200% of the score transfer threshold, resulting in the set of modified values of 5 GB/s and 20 GB/s. The intelligent agent may then include the first entity allocating 5 GB/s, allocating 10 GB/s, and allocating 20 GB/s as three possible events in the initial set of events.
The intelligent agent may use the initial set of events to simulate one or more entities that cause events using an initial set of self-play operations. A self-play operation may include using an initial set of random decisions representing actions or events caused by each entity, such as via a Monte Carlo Search Tree operation (MCTS) described further below, until a terminal program state is reached. The intelligent agent may then store the terminal program state, an associated terminal state outcome score, or other outcome program states or their associated outcome scores in the outcome program state store. The results of the initial set of self-play operations may be used to train a plurality of neural networks. The intelligent agent may perform multiple iterations of the self-play operations sequentially or in parallel for a pre-determined number of times, such as more than 10 times, more than 100 times, more than 10000 times, more than 106 times, or the like.
In some embodiments, the intelligent agent may use the results of the initial set of self-play operations to train a plurality of neural networks in combination with additional MCTS operation. The intelligent agent may begin at a first vertex of a directed graph and traverse the directed graph by proceeding to an adjacent vertex determined by the set of directed edges connected to the first vertex. Each configuration of the directed graph may be of a different program state or be otherwise associated with a different program state. For example, the intelligent agent may traverse to a first child vertex of an initial directed graph from a starting vertex. The intelligent agent may proceed to expand the directed graph to simulate evolution of its associated program state until arriving at a terminal child vertex and ending at a terminal program state. In some embodiments, performing an MCTS operation may include determining a maximum upper confidence bound (UCB) score, where the UCB score may be based on an exploration weight and a total weight score, where the exploration weight is correlated with exploring less-visited nodes and the total weight score is correlated with following the vertices associated with the greatest total reward values.
The MCTS operation may include keeping track of the number times a vertex is visited, which entity that is responsible each change in state, a prior probability value correlated with the probability of the vertex being visited in an iteration of a simulated state evolution, a backfilled vertex sum value, a hidden state value, and a predicted total reward value for an entity, or the like. After each state change, the intelligent agent may use a first neural network to determine an internal predicted value and hidden state value associated with the vertex. The intelligent agent may also use a second neural network to determine a policy weight based on the hidden state value. The intelligent agent may then expand the directed graph again, if possible, and update the policy prior value with a new policy weight using the third neural network. The internal predicted value and hidden state values of the vertices of the expanded directed graph may then be backpropagated up the directed graph to a root vertex of the directed graph.
In some embodiments, the intelligent agent may initialize each neural network of the plurality of neural networks and set a learning rate based on the number of previous training iterations. As described above, the first neural network may be trained to generate a set of internal predicted values and hidden state values, forming an internal representation of a program state. The second neural network may then be trained to formulate a set of policy weights for a policy function based on the internal representation. A third neural network may then be trained to predict a network-predicted reward value based on the policy weights and internal predicted value, where the action value may be equal to or otherwise based on the network-predicted reward value.
In some embodiments, the system may determine a set of possible state-changing events from a first vertex using one of the neural networks described above, where more than one event may cause a same adjacent vertex to be selected on the directed graph. In some embodiments, the intelligent agent may generate a decision tree associated with but not identical to the directed graph. For example, the intelligent agent may generate a decision tree having one or more decision tree vertices for each smart contract directed graph vertex, where each respective decision tree vertex associated with a respective smart contract directed graph vertex is associated with a different event or action that would trigger the respective smart contract directed graph vertex.
In some embodiments, the efficiency of tree traversal during a set of MCTS operations may be increased by limiting the outcomes based on the set of vertex categories assigned to each vertex. For example, a system may encounter a vertex associated with a category value indicating that the vertex represents or is otherwise associated with an obligation norm. In some embodiments, an obligation norm may be categorized based on a set of types of statuses, set of behaviors, or set of properties, where a status change to a satisfied status may result in the activation of a first set of child vertices and a status change to a failed status may result in the activation of a second set of child vertices. For example, a set of conditions encoded in an obligation norm associated with a vertex may be satisfied, triggering the vertex by changing the status of the vertex (“vertex status”) to a satisfied status and activating a first set of child vertices connected to the vertex via a first set of graph edges. In addition, if the obligation norm associated with a vertex is not satisfied before a failure threshold is satisfied, the system may trigger the vertex by changing the vertex status to a failed status and activate a second set of child vertices connected to the vertex via a second set of graph edges. In some embodiments, one or more obligation norms may be set to an unsatisfied status, where an unsatisfied status may indicate that the one or more obligation norms have not been satisfied but are still satisfiable.
In some embodiments, separate sets of events may be determined based on entities. For example, a first entity may be set as a first-acting entity and a second entity may be set as a counterparty entity, where each of the first-acting entity and counterparty entity may be able to cause different types of events. For example, a first-acting entity may cause events that satisfy a set of first-acting entity conditions that includes allocating different amounts of resources at different times, recalling allocated resources, or modifying a property of an allocated resource. Similarly, the resource-using entity may be able to cause one or more of a second set of events to trigger a second set of conditions associated with a second set of vertices triggerable by events causable by the counterparty entity (“set of counterparty conditions”). For example, the set of counterparty conditions may include verifying that an allocated resource satisfies a set of resource properties, confirming resource delivery, providing a digital asset after confirming the allocation of a resource, or the like.
As described above, when implementing a set of MCTS operations, the system may determine a next possible state s+1 for each state s based on a set of heuristic values associated with the set of events causable by an entity. A heuristic value may be determined using a function based on a ratio of an aggregate score associated with an event caused by an entity, the number of times that the event has been performed while performing the set of MCTS operations, a total number of iterations, and an exploration parameter. For example, the heuristic value may be determined as the sum of a first value and a second value. The first value may be a ratio of the aggregate score value to the number of times that an action has been taken while performing the set of MCTS operations. The second value may be a product of the exploration constant and a ratio of a logarithm of the total number of iterations performed while performing the set of MCTS operations and the total number of iterations. In some embodiments, each heuristic value may be associated with an event causable by an entity and may be used to determine which event is most probable. For example, if a first heuristic value is greater than a second heuristic value, an event associated with the first heuristic value may have a greater probability of being selected than an event associated with the second heuristic value.
Alternatively, or in addition, the heuristic value may be based on the vertex categories assigned to each respective vertex activated by the respective action. For example, if a first event is associated with a failure to satisfy an obligation norm, the system may automatically deduct a pre-determined failure penalty from an associated first heuristic value or set the first heuristic value to a pre-determined value. Similarly, if a second event satisfies a rights norm, the system may automatically increase the associated second heuristic value by a pre-determined amount from or set the second heuristic value to a pre-determined value. By using the vertex categories implemented in a symbolic AI model, the system may increase the efficiency and accuracy of operations to determine outcome program states based on a directed graph of an initial program state.
In some embodiments, the process 1500 may include determining a set of outcome program states or set of outcome scores or other parameters of the intelligent agent based on the set of action values or other set of parameters of the intelligent agent, as indicated by block 1528. In some embodiments, the intelligent agent may determine an outcome program state by providing a probability distribution associated with one or more possible outcome program states. For example, the intelligent agent may use a trained neural network described above to determine that there is a 25% chance that a smart contract program will end at a first terminal program state, a 35% chance that a smart contract program will end at a second terminal program state, and a 40% chance that a smart contract program will end at a third terminal program state. In some embodiments, an outcome program state may be associated with an outcome score, where the outcome score may be based on a total reward value, a total reward range, a risk of failure value, or the like. In some embodiments, a smart contract program may receive an event associated with an action performed by a counter entity that causes a change to a program state into a subsequent actual outcome program state that is not in the set of predicted outcome program states.
In some embodiments, a set of expected outcome program states determined using the operations above may be used as part of an unexpected event threshold. The system may determine that an unexpected event threshold is satisfied if an actual outcome program state is not in the set of expected outcome program states. In response, the system may perform an action such as transmitting a message indicating that an unexpected event had occurred. For example, an intelligent agent may determine that the set of expected outcome program states includes a first program state and second program state. The first example program state may be associated with a counterparty entity satisfying an obligation norm by transferring a digital asset required to access a restricted database, and the second example program state may be associated with the counterparty entity satisfying a right to rescind the request for access. If the counterparty entity instead satisfies a rights norm to override the security request, which results in a third example program state that is not in the set of expected outcome program states, the system may determine that an unexpected event threshold is satisfied. In response, the system may transmit a message indicating that an unexpected event threshold was satisfied. Furthermore, in some embodiments, the system may determine a behavior pattern of an entity based on the events caused by the entity. In some embodiments, the behavior pattern may be analyzed using a neural network to determine anomalous behavior indicating that the entity is likely to fail a future obligation or cancel the smart contract based on past behavior of the same entity or similar entities.
The set of smart contract programs may be stored on a single set of computing devices. Alternatively, one or more of the set of smart contract programs may be stored on different computing devices. The system may determine the vertices of a transaction graph based on one or more entities stored in an entity list of a smart contract program. For example, a system may access a set of five smart contract programs that include a total of seventy-five entities and generate a transaction graph having seventy-five vertices, where each respective transaction graph vertex of the transaction graph is associated with one of the seventy-five entities. In some embodiments, each vertex of the transaction graph may also have an associated smart contract list, where the associated smart contract list includes identifiers for each smart contract program that lists the entity. For example, a first vertex may have an associated smart contract list comprising a first smart contract program identifier and a second smart contract program identifier.
In some embodiments, the process 1600 may include determining a set of the transaction graph edges based on the set of smart contract programs, as indicated by block 1612. The set of transaction graph edges may represent categorical or quantitative relationships between the entities represented by the transaction graph vertices. In some embodiments, the system may expand the directed graphs of the set of smart contract programs (or other self-executing protocols) and determine their associated outcome program states to determine possible transactions between the different entities of the transaction graph. For example, the system may expand a smart contract directed graph and review each norm vertex of the smart contract directed graph to detect a first score change between a first entity and a second entity, a second score change between the first entity and a third entity, and a third score change between the second entity and the third entity. The system may update a first transaction graph edge of a transaction graph between the first entity and the second entity based on the first score change, a second transaction graph edge between the first entity and the third entity based on the second score change, and a third transaction graph edge between the second entity and the third entity based on the third score change.
In some embodiments, the system may traverse a set of directed graphs of a set of smart contract programs to modify the value associated with a transaction graph edge based on transactions between entities indicated by the directed graph vertices. For example, the system may traverse a set of directed graphs of program states, where the set of directed graphs includes a first smart contract program directed graph and a second smart contract program directed graph. If the first smart contract program directed graph includes a score change based on a condition requiring that a first entity allocates 30 terabytes of memory to a second entity, and the second smart contract program directed graph includes a score change based on a condition requiring that the second entity allocate 35 terabytes of memory to the first entity, the transaction graph edge may be updated after a system traverses through both smart contract program directed graphs. The update may change a set of score changes associated with the transaction graph edge to indicate a transfer of five terabytes from the second entity to the first entity.
Furthermore, in some embodiments, each respective smart contract program may be associated with a respective contribution weight for each entity in the transaction graph, where the respective contribution weight may indicate the proportional weight that the smart contract program has on a score change associated with the entity. For example, if a first smart contract program increases the score change for an entity by 5 and a second smart contract program increases the score change for the entity by 10, the first smart contract program may be associated with a first contribution weight of 33.3% for the entity and the second smart contract may be associated with a second contribution weight of 66.7% for the entity. In some embodiments, the set of contribution weights may be used to determine which smart contracts have had a greatest past impact on a set of entities or which are anticipated to have the greatest future impact on the set of entities.
In some embodiments, the process 1600 may include obtaining a first entity and second entity for transaction scoring, as indicated by block 1628. In some embodiments, the first entity and second entity may be explicitly provided by a user as inputs or selected by default. For example, a user signed in as the first entity or otherwise representing the first entity may use a graphic user interface and select a second entity. Alternatively, or in addition, the system may be instructed to generate a list of transaction scores and, in response, systematically perform one or more operations described in this disclosure for each of a selected set of pairs of entities, where the first and second entities are one of the pair of entities in the selected set of pairs of entities.
In some embodiments, the process 1600 may include determining a set of transaction paths between a first entity and a second entity based on the transaction graph, as indicated by block 1636. A transaction path may include a directed sequence of transaction graph edges starting at a first vertex and ending at another vertex, where each transaction graph edge may represent a transaction. For example, if a first transaction graph edge is directed from a first entity to a second entity, a second transaction graph edge is directed from the first entity to a third entity, and a third transaction graph edge is directed from the second entity to the third entity, a first path from the first entity to the third entity may include the first transaction graph edge, and a second path from the first entity to the third entity may include the second transaction graph edge and the third transaction graph edge.
In some embodiments, the path may be determined using one or more path search criteria. For example, a path search criterion may include a criterion that the path between a first entity and a second entity includes an intermediate entity, wherein each transaction graph edge of the path represents an obligation to transmit an amount of assets. In some embodiments, the path search criteria may include one or more criteria to filter out transaction paths that include transaction graph edges that are below a certain value. For example, the set of path search criteria may include a criterion that each transaction of a transaction path involves a transaction score greater than a transaction path threshold, where the transaction path threshold may be set to any value. In some embodiments, the path search criteria may be used to determine system vulnerabilities. For example, a transaction graph edge representing a token exchange between a first entity and a second entity may be mediated by a verification entity, where the first entity and second entity are both blind to each other's identities, but a failure to satisfy an obligation norm on the part of the second entity may prevent the first entity from securing an asset from the third entity. The system may determine a transaction path that includes a set of transaction graph edges to determine an exposure to failure experienced by the first entity with respect to an action or inaction of the second entity.
In some embodiments, the process 1600 may include determining an inter-entity score between a first entity and a second entity based on the set of transaction paths, as indicated by block 1642. The inter-entity score may represent one of various types of metrics and may be used to determine an entity reputation score. In some embodiments, the inter-entity score may indicate a vulnerability score, where previously-undetected vulnerabilities based on relationships between a first entity and a second entity are detected by including a set of transaction paths that include additional entities as intermediate transaction graph vertices of a transaction graph. In some embodiments, the vulnerability score may also be based on observable states of an entity. For example, a first, second, and third entity may represent a web application, a hosting application relied upon by the web application, and a cloud-connected server supplying memory resources to the hosting application, respectively. The system may assign a vulnerability score to quantify the vulnerability of the first entity to a failure of the second entity based on a net amount of computer memory the third entity is to allocate to the hosting application via an obligation norm of the cloud-connected server. The system may then determine an inter-entity score between the first entity and the third entity based on this vulnerability score.
In some embodiments, the inter-entity score may be used to update an action value. For example, some embodiments may determine an action value based on a set of reward values that includes a first reward value, where the first reward value is based on a condition requiring that a first entity transfer a score amount to a second entity. The first entity may have an inter-entity score with respect to a third entity, where the inter-entity score exceeds an inter-entity score threshold. In response to the inter-entity score exceeds an inter-entity score threshold, the first entity may modify the first reward value, such as by reducing the reward value. Alternatively, or in addition, the system may modify the first reward value based on a detected change in the status of the third entity. For example, if it is determined that the third entity is in a state that indicates it is unable to satisfy a set of obligation norms with respect to the first entity, the system may reduce the first reward value. In some embodiments, changing the reward value may be directly based on an entity reputation score, and the entity reputation score may be based on one or more inter-entity scores.
In some embodiments, the inter-entity score may indicate the detection of a cyclical path that begins and ends at a same entity. For example, the system may determine that a transaction path is cyclical, where a first entity is obligated to transfer a first amount of assets to a second entity, the second entity is obligated to transfer the first amount of assets to the third entity, and the third entity s obligated to transfer the first amount of assets back to first entity. A system may determine a cyclical path based on this relationship by detecting that the transaction path representing this relationship may have a same vertex as the start and end of the transaction path with a same or similar amount for each transaction graph edge of the transaction path. An inter-entity score may be determined based on this path based on an average amount of assets to be transferred along the cycle and include or otherwise be associated with an identifier indicating that the inter-entity score reflects a cyclical path.
In some embodiments, the system may determine that an entity is associated with a transaction that includes an obligation norm having a status indicating failure and, in response, reduces a reward associated with other possible transactions involving the entity. For example, a transaction graph may include a first entity that is scheduled to transfer a score value to a second entity at a future time, where the score value being transferred may be used to determine a reward value such as the reward values described above. The system may detect that the first entity had failed a previous obligation to transfer a different score value to a different entity and, in response, reduce the reward value based on historical data indicating that the first entity has a reduced probability of satisfying its obligation to other entities.
In some embodiments, a smart contract or other symbolic AI system may include a set of scores representing one or more properties of an entity. A smart contract may include an entity with a set of associated entity score values representing various properties of the entity. For example, a smart contract may include an entity representing a power facility. The entity may include or be associated with a first entity score indicating a maximum available power, a second entity score indicating an amount of transferable power, a third score indicating a facility fuel consumption rate, and a fourth score indicating a total failure risk.
In some embodiments, an entity may include or be associated with additional entity properties. For example, a smart contract may include an entity representing a first asset. The first entity may include or otherwise be associated with a first entity property that includes a first list of indicators, where each of the first list of indicators points to an owner of the first asset. The asset may include or otherwise be associated with a second entity property that includes a second list of indicators, where each of the second list of indicators points to another entity owned by the first entity. In addition, the asset may also be associated with quantitative score values, such as an asset valuation, a monthly cash flow, a liability value, or the like.
In some embodiments, the value of an asset or otherwise associated with the asset may be stored in various ways. For example, a score representing the value may be directly stored in a smart contract state data instead of being included as a part of entity data. For example, a smart contract score may be stored in the knowledge list 250 of the smart contract state data 200 described above, where the smart contract score may represent the value of all entities listed assets by the smart contract state data 200.
The box 1710 shows a directed graph including a set of norm vertices 1711-1713 representing a set of consecutive obligation norms. For example, the set of norm vertices 1711-1713 may be associated with a set of consecutive obligation norms to repay a loan across three intervals of time. The box 1710 also includes a norm vertex 1720 associated with an affirmative obligation norm, where an affirmative obligation norm may be associated with an affirmative covenant on the part of a first entity to provide a data payload or digital asset at a pre-determined time, and where a failure to satisfy the conditions associated with the norm vertex 1720 may result in activation of the obligation norm of the norm vertex 1723. For example, the norm vertex 1720 may represent an affirmative obligation norm that tests whether the first entity provided user verification values to a second entity, where the verified data may include information such as application health report, a financial statement, a transaction confirmation, or the like. Failure to satisfy the conditions of the norm vertex 1720 before a failure state of the norm vertex 1720 is reached may result in activation of the norm vertex 1723, which may be associated with an obligation to transmit a message or may result in the system autonomously generating a warning message. In some embodiments, a symbolic AI model may include a plurality of affirmative obligation norms associated with affirmative covenants. For example, some embodiments may include a first norm that includes norm conditions to determine whether certified financial statements are sent from a first entity to a second entity at different times. The box 1710 also includes a norm vertex 1730 associated with a prohibition norm, where the prohibition norm may include conditions and outcomes enforcing the terms of a negative covenant. For example, the prohibition norm may encode or otherwise be associated with a norm condition to determine whether any entity made acquisitions greater than a pre-determined threshold or incur certain types of debt. In some embodiments, these norm conditions may be satisfied based on an automated report provided to an API of the smart contract. Alternatively, or in addition, these norm conditions may be satisfied based on an administrator action.
The box 1740 shows a directed graph of an outcome program state that may follow the program state shown by the directed graph of box 1710. In some embodiments, one or more operations described above for the process 1400 or the process 1500 may be used to determine an outcome program state shown by the directed graph of the box 1740, where the initial program state may be represented by the directed graph of the box 1710. For example, the intelligent agent described above may determine the outcome program states represented by the box 1740 and assign it an outcome score of 85% using one or more operations described above.
As indicated by the directed graph in the box 1740, a condition of the prohibition norm of the norm vertex 1730 may be satisfied, which may in turn cause the computer system to generate an obligation norm associated with the rights norm of the fourth norm vertex 1731. In some embodiments, the rights norm of the fourth norm vertex 1731 may encode or otherwise be associated with conditions based on whether a counterparty entity transmits a message including instructions to exercise the rights norm of the fourth norm vertex 1731. In addition, an event has triggered the norm vertex 1720, where the event is a failure to satisfy the norm vertex 1720 before a failure time threshold is satisfied. The failure of the norm associated with the norm vertex 1720 activates the obligation norm associated with the norm vertex 1723. A second event has then triggered the obligation norm associated with the norm vertex 1723 by satisfying the norm condition of the obligation norm, resulting in the satisfaction norm vertex 1735. In some embodiments, the directed graph may explicitly include the satisfaction norm vertex 1735, such as in a data structure storing the directed graph. Alternatively, or in addition, the directed graph may generate an indicator indicating that the norm vertex 1723.
The box 1770 shows a directed graph representing a program state that may follow the program state shown by the directed graph of box 1740. In some embodiments, one or more operations described above for the process 1500 or the process 1600 may be used to determine an outcome program state shown by the directed graph of the box 1740, where the initial program state may be represented by the directed graph of the box 1710. As indicated by the directed graph in the box 1770, the rights norm of the fourth norm vertex 1731 may be triggered, activating a new obligation norm associated with the norm vertex 1732. In some embodiments, the new obligation norm may include norm conditions to determine whether a first entity transmits a payment amount to the second entity. For example, the new obligation norm may determine whether the first entity transmitted the entirety of a principal payment of a loan to the second entity. An event may then either satisfy or fail the norm condition encoded by or otherwise associated with the norm vertex 1732. The satisfaction of the norm vertex 1732 may result in the activation of the satisfaction norm vertex 1735, and failure of the norm vertex 1732 may result in the activation of the failure norm vertex 1737. In some embodiments, the satisfaction norm vertex 1735 or failure norm vertex 1737 may be terminal vertices. Alternatively, in some embodiments, the norm vertex 1732 may be considered a terminal vertex for embodiments where satisfaction norm vertices or failure norm vertices are not explicitly stored in a directed graph.
As shown in the directed graph 1800, the norm vertex 1801 includes the title “ra.” The “r” in “ra” may indicate that the norm vertex 1801 represents a right norm. The “a” in “ra” may indicate that the norm vertex 1801 is currently active. Combined, the name “ra” may indicate that the norm vertex 1801 represents a rights norm that is currently active. Each of the norm vertices 1811-1814 may be set as active once the rights norm represented by the norm vertex 1801 is triggered. The “op” in the title of each of the norm vertices 1811-1814 may indicate that each of the norm vertices 1811-1814 represent obligation norms which are set as active upon triggering the norm vertex 1801. Each of the norm vertices 1822-1835, 1841-1853, 1855, 1861, and 1862 may be associated with a different norm. Each of the norm vertices that include a “s” in their title represents a satisfied norm. For example, the norm vertex 1848 represents a satisfied norm, which indicates that the rights norm represented by the norm vertex 1831 is satisfied. Each of the norm vertices that include a “c” in their label represents a cancelled norm. For example, the norm vertex 1845 represents a cancelled norm, which indicates that the rights norm represented by the norm vertex 1827 is cancelled. Each of the norm vertices that include a “f” in their label represents a failed norm. For example, the norm vertex 1845 represents a cancelled norm, which indicates that the rights norm represented by the norm vertex 1827 is cancelled.
In some embodiments, each of these norm vertices may have associated data that may indicate a smart contract state at that position and may be used to determine reward values. For example, a system may traverse the directed graph 1800 by first simulating a first entity initializing a smart contract represented by the directed graph 1800 by triggering the norm represented by the norm vertex 1801 to activate the obligation norm represented by the norm vertex 1812. The computer system may then simulate a second entity not satisfying the first obligation, resulting in the activation of the norms represented by the norm vertex 1825 and 1827. The computer system may then simulate proceeding to the norm represented by norm vertex 1844, resulting in the cancellation of rights obligation represented by the norm vertex 1825 and the cancellation of the smart contract as a whole. The computer system may traverse a set of possible paths allowed by the edges of the directed graph 1800 to determine score changes for each entity and use the score changes to determine reward values associated with events, actions, vertices, or program states for one or more entities.
In some embodiments, each of the norm vertices 1841-1855 may represent a terminal norm, such as a cancelled norm or a satisfied norm. One or more events may trigger a set of norms to activate one or more new norms. For example, the satisfied norm represented by the norm vertex 1855 may be satisfied by any one of events represented by v1, v2, v3, or v4 occurring. Furthermore, a norm may be triggered by an event to activate a plurality of other norms. For example, the occurrence of the event indicated by the symbol “˜p5” may trigger the norm vertex 1811 to activate the norm vertices 1822-1823, where the event “˜p5” may indicate that the event “p5” did not occur before a failure time threshold encoded by or otherwise associated with the norm vertex 1811 is satisfied.
As indicated by the arrow 1981 pointing from the box 1980 to the box 1982, the smart contract SCA-2 represented by the box 1980 may be used to generate the smart contract SCA-2.1, which may be labeled as a child smart contract of SCA-2. Similarly, the arrow 1983 pointing from the box 1982 to the box 1984 may indicate that SCA-2.1 may be used to generate the smart contract SCA-2.1.1 represented by the box 1984. As indicated by a comparison of the score values displayed by the boxes 1980-1984, the score values associated with an entity may remain unchanged when a parent contract is deprecated and a child smart contract is initiated.
As indicated by the arrow 1915 pointing from the box 1910 to the box 1920, the smart contract SCA-1 represented by the box 1920 may be used to generate the smart contract SCA-1.1, which may be indicated to be a child smart contract of SCA-1. Similarly, the arrow 1970 pointing from the box 1910 to the box 1971 may indicate that SCA-1 may also be used to generate the smart contract SCA-1.2 represented by the box 1971. As shown by the score values displayed in the boxes 1910, 1920, and 1971, the score value 600 displayed by the box 1910 may be distributed into the score value 440 associated with SCA-1.1 and the score value 150 associated with SCA-1.2, where a score change of 10 may occur via a transaction between Entity A to another entity during the generation of the smart contract SCA-1.1.
The smart contract SCA-1.1 may then be used to generate the smart contract SCA-1.1.1, SCA-1.1.2, and SCA-1.1.3, as indicated by boxes 1922, 1931, and 1941, respectively. The score value of 440 displayed by the box 1920 and associated with SCA-1.1 may be distributed across the child smart contracts of SCA-1.1. For example, as shown by the boxes 1922, 1931, and 1941, the score value 440 associated with smart contract SCA-1.1 may be divided into the score values 20, 30, and 380, respectively. Furthermore, as indicated by the arrow 1921, the score assigned to SCA-1.1.1 is reduced by 10 during the generation of SCA-1.1.1, which reduces the score associated with SCA-1.1.1 from 30 to 20.
In some embodiments, each of the newly-generated contracts may be integrated into an object or database of the first contract SCA. Alternatively, either or both of the newly-generated smart contracts may be independent of the first contract, where each respective smart contract may or may not have a link associating the child smart contract with the parent smart contract. For example, the smart contract SCA-1.1 may include a saved property titled “parent rollover smart contract” with a field value set as to “SCA.”
In some embodiments, a party may be an entity of a child smart contract even if it is not an entity of a parent smart contract. For example, entity B may be the initial entity to which entity A is obligated to provide a total data payload equal to 760 terabytes. After entity B exercises a rights norm to generate SCA-1 and SCA-2, entity A may be obligated to allocate 600 terabytes of data for entity B and 160 terabytes of data for a new entity C. Entity B may then further exercise a rights norm to generate SCA-1.1 and SCA-1.2 to offload memory allocation instructions, where SCA-1.1 includes an obligation norm for entity A to allocate 440 terabytes to entity B and allocate 150 terabytes for a new entity D. In addition, potentially as a condition of generating additional memory allocation smart contracts on behalf of entity B, the amount of memory that entity A is to allocate for entity B may be reduced by 10.
In some embodiments, the directionality of a graph edge of a directional graph may be used to indicate a cause and outcome. For example, the tail vertex of a graph edge may be associated with a first norm and the head vertex of the graph edge may be associated with a second norm that is activated the first norm is triggered. In some embodiments, the relationship between triggering and triggered norm may be reversed, where the head vertex of a graph edge is associated with a first norm and the tail vertex of the graph edge is associated with a second norm that is activated the first norm is triggered. The directionality of a graph edge as used herein is described for illustrative purposes and may be used in different embodiments to denote different relationships between a triggering vertex and its outcome vertex. Also as used herein, the use of the article “a,” with respect an object does not necessitate a new instantiation or version of the object. For example, a directed graph may expand over time from having 1000 vertices and 800 edges to having 2000 vertices and 1800 edges and can be referred to either “the” directed graph and “a” directed graph without any ambiguity. In addition, a data element, data object, data type, or data structure such as a graph does not need to called a specific type in order to be considered a category associated with that type, and may be considered to be of that type if it includes elements or programmed relationships of that type. For example, a graph does not need to labeled as a directed graph to be considered as a directed graph, and a set of elements stored in a computer memory labeled “Terms Network” or some other title may be considered to be a directed graph if it includes components defining a directed graph, such as vertex indices representing graph vertices and a set of ordered pairs of vertex indices indicating edges of the directed graph.
As described above, some embodiments may track or predict graph evolution over time using a set of predictive operations based on the state of a smart contract program. Some embodiments may provide mechanisms to update the state in-execution program in response to one or more predicted outcomes. Some embodiments may perform operations, such as those described further below, to update or otherwise modify an in-execution smart contract program.
Modification of in-Execution Smart Contract Programs
Modifying the functions of a smart contract program or other symbolic AI program after the program has begun executing may pose challenges in distributed computing environments. For example, a set of criteria of a distributed computing platform and the time required to transfer the data needed to implement a modification in each local memory of a distributed computing platform may significantly increase the cost of operating a smart contract program on a distributed computing platform. While such computational costs may be advantageous by increasing the security of a transaction and making tampering attempts evident, they may inhibit the responsiveness of smart contract programs or other symbolic AI programs operating on a distributed computing platform. For example, some distributed computing platforms may require more than one minute, more than five minutes, or more than ten minutes to verify and distribute an update to a smart contract program. The concurrent execution of more than 10, more than 100, more than 1000, or more than 100,000 operations to amend smart contract programs on a distributed computing platform may cause network or computing performance losses. Such network or computing performance losses may reduce the reliability of smart contract programs or make them less responsive to future events. Operations or related systems that reduce the cost of amending a smart contract program may increase the responsiveness of a smart contract program to future events or verifiable changes.
Some embodiments may obtain an amendment request and extract one or more values from the amendment request usable to update a smart contract program state or other symbolic AI program state. The values extracted from the amendment request may include a set of conditional statement parameters, a set of entity identifiers, a set of conditional statement identifiers, or the like. The values extracted from the amendment request may be used to select or update a set of target norm vertices of a smart contract program directed graph. Some embodiments may determine, based on the amendment request, one or more types operations to perform such as updating a set of conditional statements, updating a set of norm vertices, updating a set of entities, or the like, where updating a set may include generating, modifying, or deleting an element of the set.
Some embodiments may simulate the updating of a directed graph based on one or more conditional statement identifiers encoded in the amendment request and modify an amendment request based on the simulation. For example, some embodiments may simulate a state change caused by an amendment request and then determine an outcome program state based on the simulated program state change. Based on a comparison of an outcome program state value with a threshold (e.g., a threshold provided by a verification agent), some embodiments may prevent the amendment request from changing the program state or modify instructions of the amendment request to satisfy the threshold.
Some embodiments may use a simulation of a state change caused by an amendment request to determine whether a modified smart contract program state would satisfy a set of criteria of one or more of the entities of a smart contract program, where the directed graph of the smart contract program is stored in a deserialized form on a first computing device. In response to a determination that the set of criteria is satisfied, some embodiments may modify the smart contract program state based on the amendment request and serialize the deserialized directed graph into a serialized array for distribution to other computing devices of a distributed computing platform. By deserializing and reserializing a directed graph based on an amendment request, some embodiments may reduce the memory needs of each computing device of the distributed computing platform. Furthermore, such operations may allow some embodiments may reduce the load on a network used to operate a distributed computing platform. However, while some embodiments may perform one or more of the above-recited operations, these operations are not necessary for some embodiments, and some embodiments may forego such operations to reduce processor use or to enjoy other advantages.
Some embodiments may select a set of entities based on the set of target norm vertices and determine whether a set of criteria of each selected entity of the set of selected entities is satisfied. Some embodiments may use the modified smart contract program state described above to determine whether the set of criteria are satisfied. Some embodiments may send a message to each respective entity of the set of selected entities, where the message may indicate that the respective entity is a participant of a target norm vertex. Alternatively, or in addition, some embodiments may include a requirement that a confirmation message authenticating the acceptance of the amendment request from each respective entity of the set of selected entities is obtained.
Some embodiments may update a smart contract program state in response to determining that the set of criteria are satisfied, such as by updating a set of conditional statements, updating their associated target norm vertices, updating their associated entities, or the like. Some amendment requests may further cause some embodiments to associate a newly-generated directed graph portion or an existing directed graph portion to one or more target norm vertices. Some embodiments may assign priority category values to one or more norm vertices, where the priority category values may be used to determine an order of vertex activation or vertex triggering in the case of a single event triggering multiple norm vertices.
As indicated by the first directed graph edge 2013, failing to satisfy the conditional statement of the first norm vertex 2012 may result in the activation of the second norm vertex 2014, which may be categorized as a rights norm. The rights norm represented by the second norm vertex 2014 may represent a right of the first entity “Ent1” to allocate 105 units to the second entity “Ent2.” In some embodiments, while not shown, satisfaction of the first norm vertex 2012 or second norm vertex 2014 may cause the activation of a norm vertex having no further child vertices that indicate that the obligation is fulfilled. Alternatively, or in addition, some embodiments may update a label to indicate that the norm vertex satisfied and has no additional child vertices. As indicated by the second directed graph edge 2015, failing to satisfy the conditional statement of the first norm vertex 2012 may also result in the activation of a third norm vertex 2016 categorized a rights norm. The third norm vertex 2016 may represent a right of the second entity “Ent2” to force the allocation of an amount represented by the value “TTL” by the first entity “Ent1,” where “TTL” may be variable number dependent on a remaining amount. For example, the third norm vertex 2016 may represent a right of the second entity “Ent2” to force the allocation of 100 GB of a non-duplicable resource by the first entity “Ent1.”
The smart contract program state associated with the directed graph 2010 may be modified by the amendment request 2030. The amendment request 2030 may include various parameters. For example, the amendment request 2030 may include a date on which the amendment is to take effect, a conditional statement identifier indicating the conditional statement that the amendment request 2030 is to modify or replace, or a pair of modifying conditional statements. The first modifying conditional statement of the pair of modifying conditional statements may indicate that the first entity “Ent1” is to transfer 10 units to a third entity “Ent3.” A second modifying conditional statement of the pair of modifying conditional statements may indicate that the first entity is to transfer 89 units to the second entity “Ent2.” As discussed further below, some embodiments may extract conditional statement parameters such as the entity identifiers, the quantitative amounts, the date, and the conditional statement identifier.
The directed graph 2040 may be an outcome directed graph after a state change to the program state of the directed graph 2010, where the state change is caused by the amendment request 2030. The directed graph 2040 includes a first directed graph portion 2050 and a second directed graph portion 2060, where the first directed graph portion 2050 and the second directed graph portion 2060 are disconnected from each other. The first directed graph portion 2050 includes the first norm vertex 2012, where the entity identifier “Ent3” replaces the entity identifier “Ent2” for the entity field “P2.” In some embodiments, conditional statements of adjacent norm vertices may be affected based on the amendment request, even if the adjacent norm vertex is not directly referenced by an amendment request or does not use a conditional statement identified by an amendment request. For example, the second norm vertex 2014 may be updated such that the quantity “105” is changed to the quantity “94” based on the quantity of the rights norm the set as equal to the sum of the first quantity and an additional five units. Additionally, the third norm vertex 2016 may be updated such that the quantity “TTL” is changed to the quantity “(TTL-10).” Furthermore, some embodiments may include a score change, transfer of scores, or allocation of resources as a result of modifying a smart contract program based on the amendment. For example, an amendment request may cause some embodiments to cause the first entity to directly allocate an hour of processor time to a fourth entity.
The directed graph 2040 includes the second directed graph portion 2060, where the second directed graph portion may include a fourth norm vertex 2061, a fifth norm vertex 2063, a sixth norm vertex 2065, a seventh norm vertex 2067, an eighth norm vertex 2069, and a ninth norm vertex 2071. A conditional statement of the fourth norm vertex 2061 may include a condition that is satisfied if a first entity used to fill the entity field P1 allocates a first quantity 10 units to a second entity used to fill the entity field P2, where the first entity has the entity identifier “Ent1,” and the second entity has the entity identifier “Ent3.” As indicated by the third directed graph edge 2062, failing to satisfy the conditional statement of the fourth norm vertex 2061 may result in the activation of the fifth norm vertex 2063. The rights norm represented by the fifth norm vertex 2063 may represent a right of the first entity “Ent1” to allocate 11 units to the second entity “Ent3,” which may be interpreted as curing a failure to satisfy the fourth norm vertex 2061. As indicated by the second directed graph edge 2064, failing to satisfy the conditional statement of the fourth norm vertex 2061 may also result in a third rights norm represented by the sixth norm vertex 2065. The third rights norm represented by the sixth norm vertex 2065 may represent a right of the second entity “Ent3” to force the allocation of an amount represented by the variable TTL/10 by the first entity “Ent1.” Furthermore, as indicated by the directed graph edge 2072, failing to satisfy the conditional statement of the sixth norm vertex 2065 may result in the activation of the seventh norm vertex 2073. An obligations norm represented by the seventh norm vertex 2073 may represent an obligation of the second entity “Ent2” to allocate eight units to the third entity “Ent3.”
In some embodiments, the process 2100 of
In some embodiments, as described above, the smart contract program state may encode a directed graph in the form of a serialized array of norm vertices and its corresponding set of graph edges. For example, the smart contract program state may include a serialized array of norm vertices “[1, 4, 7],” where each number of the numbers indicates a norm vertex, and a corresponding set of directed graph edges “[[1,4], [4,7]],” where each subarray indicates a directed graph edge. Alternatively, or in addition, the smart contract program state may include a plurality of serialized arrays of norm vertices and their corresponding edges. For example, the smart contract program state may include a first serialized array of norm vertices “[1, 4, 7]” associated with a corresponding serialized array of directed graph edges [[1, 7], [7,4]]. The smart contract program state may include a second serialized array of norm vertices “[2, 5, 7, 4, 41]” and their corresponding directed graph edges [[2,5], [7,5], [5,4], [4,41]]. As discussed further below, storing portions of a graph in separate serialized arrays of norm vertices may contribute to increasing memory efficiency or update efficiency when executing a smart contract program state.
In some embodiments, the directed graph may be disconnected, having two or more unconnected portions. For example, the smart contract program state may include a first serialized array of norm vertices “[1, 4, 7]” associated with a corresponding serialized array of directed graph edges [[1, 7],[7,4]]. The smart contract program state may also include a second serialized array of norm vertices “[2, 5, 7, 4, 41]” and their corresponding directed graph edges [[2,5], [4,5], [4,41]], where the graph portion formed by the first serialized array of norm vertices is disconnected from the graph portion formed by the second serialized array of norm vertices.
In some embodiments, the process 2100 may include obtaining an amendment request encoding a set of conditional statement parameters, as indicated by block 2108. The amendment request may be obtained at an API of a smart contract program, API of application in communication with the smart contract program, an API of a distributed computing platform, an API of a computing system executing the smart contract program, or the like. Some embodiments may obtain the amendment request from data entered by a user at a graphical user interface. Alternatively, or in addition, some embodiments may obtain an amendment request that was machine-generated or updated using a machine-learning system. The set of conditional statement parameters may include various types of information, such as a date of enforcement, a quantitative value, a categorical value, an entity identifier or other identifier, or the like. For example, an amendment request corresponding to the natural language instructions “entity 12591xc3 is obligated to allocate 300 GB to entity 27831t6” may include the quantity parameter “300” and the entity identifiers “12591xc3” and “27831t6.”
The set of conditional statement parameters may include some or all of a conditional statement written in a computer-readable programming language. For example, the set of conditional statement parameters may include the conditional statement “if (ENTITY==“entity 1×1”): SENDRESOURCE(100, “entity 1×1”, “entity 1×2”)”. Alternatively, or in addition, the set of conditional statement parameters may include a set of parameter values used to fill a field a conditional statement. For example, the set of conditional statement parameters may include a conditional statement identifier “cond_state10105421” with the associated parameters [“entity1x1”; “entity1x2”; 194]. The conditional statement identifier “cond_state10105421” may identify a conditional statement that may be represented in the form “if RCVD_FUNCTION(ARG1, ARG2, ARG3).” The function “RCVD_FUNCTION” may be a function that accepts the three parameters ARG1, ARG2, and ARG3 and returns the boolean “true” if ARG1 and ARG2 are entities and ARG1 has received the amount ARG3 from the entity identified by ARG2. The associated parameters may indicate that “entity1x1” is to be used in place of ARG1, that “entity 1x2” is to be used in place of ARG2, and that “194” is to be used in place of ARG3.
In some embodiments, the amendment request may include specific instructions to generate a new directed graph portion. For example, the amendment request may include the instructions “Generate_Vertex(1525, 215, “satisfied,” 15216),” which may cause a smart contract program to generate a norm vertex identified as “1525.” A graph edge may associate the newly-generated norm vertex to the norm vertex identified as “215” and be activated based on the satisfaction of the norm vertex identified as “215.” The final parameter of the function “Generate_Vertex” may indicate that the newly-generated norm vertex may be associated with a conditional statement identified as “15216.” Alternatively, or in addition, some embodiments may determine that an amendment request includes instructions or values satisfiable by the generation of a new directed graph portion and, in response, generate the new graph portion. For example, some embodiments may determine that the instructions include “if vert_failed(vert[1653]): cond_state[15216],” where the instructions may cause a smart contract program to determine that a norm vertex associated with the conditional statement “15216” is in existence.
In some embodiments, the process 2100 may include determining a set of target norm vertices of the smart contract program state or other symbolic AI system based on the amendment request, as indicated by block 2116. In some embodiments, the amendment request may include direct references to one or more norm vertices of a directed graph of a smart contract program. For example, the event request may include a reference to the norm vertex “OP111-1,” where the value “OP111-1” is a norm vertex identifier for a norm vertex in a smart contract graph. Alternatively, the amendment request may include a series of the old parameters to identify one or more smart contract graph norm vertices. For example, some embodiments may receive an amendment request that includes a set of norm vertex-identifying parameters, such as a parameter that specifies a transaction date, a set of affected entities, or a transaction amount. Some embodiments may extract the conditional statement parameters or other norm vertex-identifying parameter from the amendment request, where a norm vertex-identifying parameter may be any value that can be used to identify a set of norm vertices. For example, a norm vertex-identifying parameter may include a norm vertex identifier, a category type specific to a set of norm vertices, a time or time interval that can be used to isolate a set of norm vertices having fulfillment deadlines due after the time or within the time interval, or the like. Each parameter of the set of norm vertex-identifying parameters may then be used by a computing system to determine one or more norm vertices or conditional statements. For example, an amendment request specifying a modification for all transactions associated with a conditional statement identifier after a specified date may result in the extraction of the conditional statement identifier and the specified date as norm vertex-identifying parameters.
Some embodiments may determine a set of active norm vertices of a directed graph of a smart contract program or other symbolic AI program. Some embodiments may search through the set of active norm vertices instead of searching through all norm vertices of the directed graph. In some embodiments, an active norm vertex may be a norm vertex having an associated conditional statement that may be triggered, whereupon the triggering of the associated conditional statement causes the activation of another norm vertex to take place or may otherwise cause an state change to occur (e.g., update a label indicating that an obligation norm is satisfied). For example, some embodiments may determine that a first, second, and third norm vertex are each active norm vertices of a set of active norm vertices. Some embodiments may then search through the set of active norm vertices that satisfy a set of norm vertex-identifying parameters to determine a target norm vertex. For example, an amendment request may include norm vertex-identifying parameters specifying a change to terms associated with an allocation of 100 units by a first entity for a second entity. Some embodiments may extract the first entity identifier, the second entity identifier, and the allocation of 100 units to form a set of norm vertex-identifying parameters and search through a set of active norm vertices to find which norm vertices satisfy the set norm vertex-identifying parameters. This is not to suggest that all embodiments may restrict the search to a set of active norm vertices, and some embodiments may search through other norm vertices of a directed graph or all norm vertices of a directed graph for their own benefit(s), such as for ensuring a completeness of the search or for providing data for a study of historical performance.
Some embodiments may obtain amendment requests that may modify one or more conditional statements and search for a list of affected conditional statements, which may then be used to select a target norm vertex. For example, some embodiments may include an amendment request that includes the computer-interpretable code ‘if entity[2123].sending( )==true, Replace(2123, 254121),’ which may be converted from the natural language statement, “if the entity having the entity identifier ‘2123’ is sending a resource, replace it with the entity having the entity identifier ‘254121.’” Some embodiments may then search through a set of conditional statements to determine which of the set of conditional statements are affected by the amendment request based on which entities are part of a condition of the conditional statement. For example, a first conditional statement may include the condition ‘if entity[2123].sending( )==true: message(entity[254121], “SENT”).’ Some embodiments may determine that the first conditional statement is an affected conditional statement and then determine a set of target norm vertices that includes, uses, or is otherwise associated with the first conditional statement.
In some embodiments, the process 2100 may include determining a set of selected entities based on the set of target vertices, as indicated by block 2120. As discussed above, the set of target vertices may be associated with a set of entities via the set of conditional statements encoding one or more of the set of entities. This set of entities may be used as a set of selected entities, where the set of selected entities may indicate entities that are directly affected by the amendment request. By determining which entities are directly affected by an amendment request, some embodiments may more effectively message the set of affected entities without overloading a messaging system by notifying only entities that are likely to have an interest in the amendment request. For example, a smart contract program may include a list of 100 entities, amongst which 25 are determined as being affected by an amendment request based on the entities associated with vertices affected by the amendment request. These 25 entities may then be sent a first message indicating that they are affected by the amendment request, while the other 75 entities are not sent any messages or sent a message different from the first message. Furthermore, some embodiments may require that an amendment request be confirmed by the set of selected entities instead of requiring that confirmation be provided by all entities of a smart contract program. By reducing the number of required confirmation messages, some embodiments may significantly reduce the time needed to amend a smart contract program having more than ten entities, more than twenty entities, more than two hundred entities, or the like by requiring a confirmation message from only a subset of the entities of the smart contract program.
In some embodiments, the process 2100 may include simulating a modification of the smart contract program state or other symbolic AI program state based on the amendment request, as indicated by block 2124. Some embodiments may simulate the modification of an application program state based on the amendment request by generating a version the smart contract program state and changing the version based on the amendment request without requiring the distribution of the changed version to other computing devices. Simulating a modification of the smart contract program state may include generating a new graph structure that is different from the graph structure of an unmodified smart contract program state with respect to the number of norm vertices, the number of edges, or the set of logical categories associated with each of the norm vertices. For example, a first graph structure of unmodified smart contract program state may include three active obligation norm vertices. After simulating a modification based on the amendment request, the graph structure of the modified smart contract program state may be changed such that the logical category of one norm vertex is changed from being an obligation norm to a prohibition norm. This may cause the simulated modification of the graph structure to include two active obligation norm vertices and one active prohibition norm vertex. Simulating a modification of the smart contract program state may include generating a version of a set of conditional statements and updating the version of the set of conditional statements based on the set of conditional statement parameters.
As further discussed in this disclosure, a simulated modification based on an amendment request may be used to determine whether an outcome state caused by the amendment request satisfies a set of requirements of the smart contract program. For example, as further illustrated below, an amendment request to a smart contract program state may cause a first entity to be obligated to allocate five terabytes of memory to a second entity. Some embodiments may simulate an implementation of the amendment request and determine that the simulated outcome state violates a third entity's requirement. For example, the third entity's requirement may be that the second entity is prohibited from reserving memory allocated by the first entity, where third entity's requirement may be implemented as a prohibition norm or may be implemented in another form (e.g., program-wide rule) encoded in the smart contract program. In response to this simulated violation, some embodiments may prevent the amendment request from modifying the smart contract program state or may modify the instructions of the amendment request (e.g., by changing the source entity for the allocated memory).
Some embodiments may simulate the modification of an application program state without simulating every modification caused by the amendment request. For example, a first amendment request may cause the modification of a program state to change the conditional statement and logical category of a first norm vertex. The change may cause an obligation norm of a first entity to allocate a first amount to a rights norm triggerable by a second entity to request a second amount from the first entity after an interval of time encoded in the amendment request. The computing system may simulate the modification of the application program state by changing the logical category associated with the first norm vertex and any associated graph structure changes without changing the first amount to the second amount or include the encoded interval of time. Based on a determination that a simulated modified graph structure is not identical to or otherwise different from a graph structure before a smart contract modification, some embodiments may generate a new norm vertex for a directed graph of a smart contract program.
In some embodiments, a representation of the simulated modification of the application program state based on the amendment request may be sent to one or more verification agents. A verification agent may include a third-party entity, an automated testing system, or another system. For example, some embodiments may simulate modification of a first smart contract program and send the graph structure, their corresponding conditional statements, or other related data to a verification agent for display a graphical user interface. Alternatively, some embodiments may simulate the modification of the application program state without any messages. The verification agent may send a message via an API, where the message may confirm that the simulated modification is acceptable, reject the simulated modification or may include a second simulated modification of the smart contract program state. As further described below, some embodiments may require a message confirming the authorization of the simulated modification before proceeding to use the simulated modifications for additional analysis. Alternatively, some embodiments may use the simulated modifications for further analysis without requiring a confirmation message.
Some embodiments may simulate the occurrence of a set of simulated events or a sequence of simulated events for a simulated modified smart contract program. For example, some embodiments may simulate the occurrence of a sequence of simulated events indicated to have occurred based on an associated set of occurrence times, where the associated set of occurrence times indicate the occurrence of a first simulated event of the sequence of simulated events on a first day and the occurrence of a second simulated event of the sequence of simulated events on a second day. Some embodiments may simulate the occurrence of a plurality of simulated events or a plurality of sequences of simulated events to determine a set of outcome scores corresponding to a set of simulated modified smart contract program states. For example, some embodiments may set the average amount of computing memory allocated over one month as an outcome score and determine a set of outcome program states across three equally-likely sequences of events based on five different simulated program states.
Some embodiments may determine a respective set of outcome program states for each respective simulated program state of the five different simulated program states, where each respective simulated program state is a result of modifying the program state based on a respective amendment request of the plurality of amendment requests. Some embodiments may then determine a respective set of outcome scores for each respective set of outcome program states. Some embodiments may then select an amendment request from the plurality of amendment requests based on the set of outcome scores, where the set of outcome scores may include each of the respective set of outcome scores. For example, some embodiments may select an amendment request based on which results in a maximum outcome score of the set of outcome scores. Some embodiments may perform one or more of the simulation operations described above using a single computing device or subset of computing devices of a distributed computing platform instead of using each computing device of the distributed computing platform to simulate a modification. By performing the simulation on a one computing device or a small number computing devices, some embodiments may reduce the computational load on the distributed computing platform and reduce the network traffic used to operate the distributed computing platform.
In some embodiments, the process 2100 may include determining whether the amendment request satisfies the set of criteria of the set of selected entities, as indicated by block 2128. Determining whether the amendment request satisfy the set of criteria of the set of selected entities may include determining whether the amendment request satisfies a set of governing conditional statements associated with the smart contract program. For example, a governing set of conditional statements may prohibit transactions with entities of a first entity type, and the amendment request may change a participating entity to an entity of the first entity type. In response, some embodiments may determine that the amendment request does not satisfy the set of criteria of the set of selected entities. In some embodiments, each entity of the set of selected entities may have a different set of criteria that must be satisfied in order to provide a respective confirmation message.
In some embodiments, the set of confirmation messages may include one or more authentication frameworks to authenticate a confirmation message. For example, the set of confirmation messages may include a set of passkey values. For example, each respective message of a set of confirmation messages may include a respective passkey value of the set of passkey values, where each message of the set of confirmation messages may be associated with a respective entity of the set of selected entities. For example, an entity may send a confirmation message, including a human-entered or machine-provided passkey value. Some embodiments may compare a respective passkey value with a respective stored passkey value to determine whether the respective passkey value matches with the respective stored passkey value. In some embodiments, some embodiments, the passkey value may be encrypted and compared to a set of encrypted passkey values to determine a match. Alternatively, or in addition, a passkey value may be decrypted and compared to a set of decrypted passkey values to determine a match. Based on the entity-sent passkey value matching a stored passkey value in a set of stored passkey value, some embodiments determine that a criterion associated with one or more of the set of selected entities is satisfied. While the above describes an implementation of one type of authentication framework, some embodiments may use one of various other types of authentication frameworks when sending confirmation messages or other messages. For example, some embodiments may implement a Public Key Infrastructure (PKI) framework, such as that described in “Introduction to public key technology and the federal PKI infrastructure” (Kuhn, D. Richard, et al. National Inst of Standards and Technology Gaithersburg MID, 2001), which is hereby incorporated by reference. Some embodiments may use various data transport protocols when implementing the authentication framework, such as secure socket layer (SSL) or transport layer security (TLS).
Various types of criteria may be used to determine whether to modify a smart contract program state based on the amendment request. In some embodiments, determining whether a criterion is satisfied may include determining whether an entity of a smart contract program is one of a set of prohibited entities or one of a set of prohibited entity types. For example, the entity “entity1” may have the entity type “x1x1,” and determining whether the set of criteria is satisfied may include determining that entities of the entity type “x1x1” are entities of a prohibited entity type. In response, some embodiments may determine that the set of criteria is not satisfied by the entity “entity 1.”
In some embodiments, determining whether the set of criteria is satisfied may include determining whether a non-duplicable asset is concurrently transferred or allocated to different entities based on a single event or sequence of events. A non-duplicable asset may include an amount of computing time on a specific computing resource during a specific time interval. For example, a non-duplicable asset may include an allocated utilization time between 04:00 and 06:00 on a specified set of processor cores. The transfer or allocation of a non-duplicable resource to multiple resources may be detected as a conflict, and some embodiments may include a verification mechanism to prevent the conflict. For example, some embodiments may determine a simulated contract program state based on a simulation of the modification of a smart contract program state based on an amendment request. Some embodiments may then simulate how the simulated smart contract program state responds to a sequence of events and determine that a first entity is allocating control of a specific computing resource to a second entity based on an event, and that the first entity will also be caused to allocate control of the specific computing resource to a third entity based on the event. Some embodiments may determine this concurrent allocation and, in response, prevent the amend request from being implemented, send a message indicating that the amendment may cause a conflict, or the like.
In some embodiments, the set of criteria may include determining that an entity, set of entities, or the entity type is required for transactions of a specified transaction type or all transaction types. For example, some embodiments may include a criterion that each entity of a smart contract program is indicated as verified based on a verification field being populated with the value “verified.” As another example, the entity “entity2” may have the entity type “x2x2” and a possible criterion may be that all entities of a smart contract program state be of the entity type “x2x2.” In response, some embodiments may determine that the entity “entity2” satisfies the criterion.
Some embodiments may perform one or more of the determination operations described above using a subset of computing devices of a distributed computing platform instead of using each computing device of the distributed computing platform to determine whether a set of criteria are satisfied. For example, some embodiments may store a version of each of a set of criteria of a set of entities at a storage memory. Some embodiments may then determine whether an amendment request satisfies the set of entities using a computing device that includes or is otherwise capable of accessing the storage memory. After determining that the set of criteria is satisfied, some embodiments may send the amendment request or parameters stored in the amendment request to other computing devices of the distributed computing platform. By restricting the determination operation to one computing device or a small number of computing devices, some embodiments may reduce the overall computational load on the distributed computing platform and reduce the network traffic used to operate the distributed computing platform. In some embodiments, if the amendment request satisfies the set of criteria of the set of selected entities, operations of the process 2100 may proceed to block 2140. Otherwise, operations of the process 2100 may proceed to block 2132.
In some embodiments, the process 2100 may include updating an amendment request, as indicated by block 2132. In some embodiments, the amendment request may be updated in response to a failure to satisfy the set of criteria of the selected entities. In some embodiments, the amendment request being generated may be generated in a parameter space that allows the amendment request to have multiple possibilities. In some embodiments, some of these possibilities of the amendment request may satisfy the set of criteria of the selected entities while other possibilities of the amendment request May not satisfy the set of criteria of the selected entities. For example, a first version of the request may change a first quantity from the value “100” to the value “300” and a first criterion of one of the selected entities may require that the value of the first quantity the less than the value “200.” In response, after determining that the amendment request failed the first criterion, some embodiments may update the agreement request to change the first quantity from the value “300” to the value “150” using one or more of various types of optimization methods or machine-learning methods.
Some embodiments may send a message to one or more of the set of selected entities in response to failing to satisfy the set of criteria of the selected entities. For example, some embodiments may send a message to all of the set of selected entities in response to failing the set of criteria. Some embodiments may send a message indicating that the set of criteria has been failed without updating the amendment request. Additionally, or alternatively, some embodiments may send a message to an entity or other agent that sent the amendment request indicating that the amendment request on a program state resulting from the amendment request has failed the set of criteria.
In some embodiments, the process 2100 may update a set of conditional statements based on the set of conditional statement parameters, as indicated by block 2140. In some embodiments, determining the updated set of conditional statements may include replacing one or more conditional statements with a new conditional statement determined from the set of conditional statement parameters. For example, a norm vertex may be associated with a first conditional statement, where the first conditional statement is indicated to be replaced by a second conditional statement encoded in an amendment request. In some embodiments, a conditional statement that is to be replaced or otherwise unused may be marked as deprecated. For example, some embodiments may change a usage indicator associated with a conditional statement to indicate that the conditional statement is deprecated based on an amendment request indicating that the conditional statement should be removed from use in the smart contract program. Alternatively, instead of deprecating a conditional statement stored in a set of conditional statements, some embodiments may determine delete the conditional statements.
In some embodiments, updating the set of conditional statements may include filling out, replacing, or otherwise using one or more conditional statement parameters to populate fields of the set of conditional statements. For example, the computing system may obtain a first plurality of entity identifiers, and a user may use the entity identifiers fill out the function that uses the first plurality of entity identifiers. By filling out a field of an existing conditional statement instead of replacing the condition statement, some embodiments may increase the speed by which a smart contract program state may be distributed on a distributed computing platform. However, while the above describes filling, replacing, or otherwise using one or more conditional statement parameters to populate fields of the set of conditional statements, some embodiments may forego such operations and update the set of conditional statements using other methods.
In some embodiments, updating the set of conditional statements may include adding one or more conditional statements to the set of conditional statements. Some embodiments may determine an updated set of conditional statements indexed by conditional statement identifiers by adding a new conditional statement that may have its own associated conditional statement identifier. For example, an amendment request may include an instruction to add a specific conditional statement to a set of conditional statements of a smart contract program. In response to obtaining the amendment request, some embodiments may include instructions to add a conditional statement identifier “x1x1” to a conditional statement identifier index.
In some embodiments, updating the set of conditional statements may include determining whether a conditional statement (or an associated norm vertex) was triggered by a past event. An outcome of the conditional statement may include a transaction between a first entity and a second entity. For example, after determining that a past event had triggered a norm vertex based on an amendment request modifying the norm vertex or an associated conditional statement, some embodiments may determine a first score value associated with a transaction caused by the outcome. Some embodiments may determine a second score value encoded in the amendment request and determine a difference between the first score value and the second score value. Some embodiments may then initiate a transaction between the first entity and the second entity based on the score difference. By using a score differences to account for differences between amendment requests and past events, some embodiments may include mechanisms to retroactively apply an amendment request.
In some embodiments, updating the set of conditional statements, set of target vertices, or other values of a program state operating on a distributed computing platform may include sending a set of values to each computing device of the distributed computing platform. In some embodiments, to determine the validity of a distributed value, the smart contract program may use one or more consensus algorithms. For example, to reach a consensus on the validity of a set of conditional statements, some embodiments may use a consensus Paxos algorithm, a Raft algorithm, HotStuff, or the like. Furthermore, some embodiments may centralize one or more of the operations described above at a single computing device or a subset of computing devices and then send a processed set of values to other computing devices of the distributed computing platform. For example, some embodiments may simulate modifications of a smart contract program using a first computing device and determine an updated amendment request based on the simulated modifications before sending the updated amendment request to other computing devices. Additionally, some embodiments may store or otherwise have access to a set of criteria of each entity of a set of selected entities to determine whether the set of criteria is satisfied. By centralizing operations at a single computing device or a subset of computing devices, some embodiments may reduce the overall computational cost of amending a smart contract program or other symbolic AI program.
In some embodiments, the process 2100 may include updating the set of target norm vertices or values associated with the set of target norm vertices based on the amendment request, as indicated by block 2144. Updating the set of target norm vertices may include updating a field of the target norm vertex for a conditional statement identifier of the target norm vertex with a new conditional statement identifier. Updating the field may associate the target norm vertex with the new conditional statement having the new conditional statement identifier. For example, a first target norm vertex may have or otherwise be associated with the conditional statement identifier “4457” and updated to have instead or otherwise be associated with the conditional statement identifier “9941.” After the update, a first event satisfying the conditional statement having the identifier “9941” may trigger the first target norm vertex, whereas a second event satisfying the conditional statement having the identifier “4457” does not trigger first target norm vertex.
Some embodiments may update the smart contract program state by deserializing a serialized array representing a directed graph stored in persistent memory. The deserialized directed graph may be stored in non-persistent memory for fast processing or operations. Some embodiments may then add, modify, or remove a norm vertex or edge of the directed graph stored in the non-persistent memory and then reserialize the directed graph. For example, some embodiments may include a first serialized array “[1 3 5]” having an associated serialized array of edges “[[1,3], [3,5]]” stored in a persistent memory of a computing system. Some embodiments may deserialize the serialized array into an adjacency matrix form stored in the non-persistent memory and add a norm vertex having a norm vertex identifier “7” and edge directing from the norm vertex “5” to the norm vertex “7.” Some embodiments may reserialize the deserialized directed graph to determine the updated serialized array “[1 3 5 7]” having an associated serialized array of edges “[[1,3], [3,5], [5,7]].” The deserialization and reserialization of directed graph data may result in increased storage memory use efficiency, or network performance efficiency (e.g., in the case of the smart contract being implemented on a distributed computing platform). However, while the above suggests some embodiments may implement a deserialization/reserialization operation, such operations are not necessary. Some embodiments may forego such operations and use other methods to benefit from increased computational performance efficiencies, network performance efficiencies, or the like.
In some embodiments, the process 2100 may include updating a directed graph of a smart contract program with an additional set of norm vertices, as indicated by block 2148. Some embodiments may associate an additional set of norm vertices with a directed graph of the smart contract program in response to an amendment request causing the creation of the additional set of norm vertices. For example, some embodiments may determine that an amendment request includes instructions or values that causes the generation of a new norm vertex in the directed graph based on a determination that a set of conditional statements parameters in the amendment request is unrelated to an existing norm vertex of the directed graph. Some embodiments may then generate a norm vertex by creating a new conditional statement and associating the new conditional statement with a newly created norm vertex having associated directed graph edge. In some embodiments, the amendment request may cause the creation of a plurality of norm vertices and their corresponding directed graph edges.
In some embodiments, the generated set of norm vertices may be serialized into a serialized array in a persistent memory. In some embodiments, the generated set of norm vertices may be deserialized into a deserialized directed graph stored on a non-persistent memory, where the deserialized directed graph may include an adjacency matrix or adjacency list. Some embodiments may generate an edge that connects one or more of the set of norm vertices with one or more norm vertices of the set of target norm vertices determined above. In some embodiments, one or more of the set of generated norm vertices may have an associated priority category value used to determine a sequence by which different norm vertices are triggered. In some embodiments, the implementation of an execution sequence based on a set of associated priority category values may be used to reduce the risk of contradictions or logical errors.
Some embodiments may associate a target norm vertex with another portion of an existing directed graph based on an amendment request. The other portion of the existing directed graph may be part of the smart contract program state. Alternatively, the other portion of the existing directed graph may be part of a different smart contract program state. For example, an amendment request for a first smart contract program may include a program identifier of a second smart contract program and a vertex identifier of a first norm vertex of the second smart contract program. Some embodiments may associate a target norm vertex with the first norm vertex based on the program identifier and the vertex identifier.
Various formats may be used to indicate this cross-program relationship or the order by which active norm vertices are triggered across different smart contract programs. For example, some embodiments may add a new directed graph edge to a set of directed graph edges, where the new directed graph edge may point from the target norm vertex to the norm vertex of the other smart contract program. Some embodiments may account for potential confusion by having the directed graph vertex point to a dummy norm vertex, where the dummy norm vertex includes values identifying the norm vertex of the other smart contract program. Some embodiments may assign a priority category value to each respective norm vertex or respective smart contract program and refer to the respective priority category values to determine an order by which norm vertices are triggered in response to an event that triggers multiple norm vertices.
In some embodiments, the process 2100 may include storing the smart contract program state or other symbolic AI program state in a persistent memory, as indicated by block 2152. After updating the program state of a smart contract program or other symbolic AI model, some embodiments may then store the smart contract program in a persistent storage memory of one or more computing devices. In addition, some embodiments may store the amendment request, rejected amendment requests, or other data related to an amendment request in a same storage memory.
As described above, some embodiments may modify or otherwise update an in-execution smart contract program. Additionally, some embodiments may reduce an effective amount of data to analyze when categorizing or comparing directed graphs using dimensional reduction operation before and after amending a smart contract program. Some embodiments may perform operations, such as those described further below, to more efficiently analyze one or more types of directed graphs by reducing the amount of data used to analyze a directed graph.
Dimensional Reduction of Categorized Directed Graphs
Attempts to analyze or optimize real-world programs used to formalize, automate, or enforce repeated transactions between multiple entities may quickly suffer from the curse of dimensionality. Such programs are often challenged with large datasets having multiple (over 5, over 10, over 100) variables and more than 20, more than 50, more than 1000, or more than 1,000,000 individual records due to both broadly-applicable challenges and domain-specific challenges. Broadly used operations such as processing a large volume of data, determining similarities between different programs, and implementing meaningful visualization of relatively small sections of the data may be exacerbated by data characteristics in specific domains. In the case of smart contracts or other symbolic AI programs, domain-specific issues may include adapting operations to account for the various types of features that can be obtained from data stored in program state, relationships between different conditional statements, and differing priorities of different data consumers. The variety of data types, complex nature of relationships between terms or entities of a smart contract, and different priorities of different visualization audiences may challenge attempts to determine the most significant variables of a smart contract, compare the similarity of different contracts, and visualization attempts.
Some embodiments may use a directed graph encoding elements of a smart contracts or other symbolic AI programs to determine a set of features and vertices to prioritize or display. Some embodiments may determine a set of features for a set of vertices of the directed graph, where the features and their corresponding feature values for a vertex may be used to characterize the vertex, its relationships with other vertices in the directed graph, or the like. Some embodiments may include determining multiple candidate subsets of features or performing other feature selection operation to determine a prioritized subset of features from the set of features. Some embodiments may perform a feature extraction operation that accounts for the prioritized subset of features to determine a prioritized subset of vertices, where such operations may include using a neural network, applying principal component analysis (PCA), or the like. Some embodiments may then visually indicate the prioritized subset of features or the prioritized subset of vertices. Some embodiments may apply dimension reduction results to compare different graphs to determine which elements of a program state may be common across multiple smart contract programs and, as a result, also determine which set of vertices are uncommon.
By visualizing smart contract programs or other symbolic AI programs encoding transactions between multiple entities as directed graphs and indicating outcome-relevant vertices, some embodiments may increase the interpretability of directed graphs. Furthermore, some embodiments may use category labels associated with a vertex or other structural associated with a conditional statement to increase the efficiency and accuracy of analysis results. For example, as further discussed below, a feature of a vertex may include the category label “right,” which is selected from a set of mutually exclusive categories that may include the category labels “right,” “obligation,” or “prohibition.” Such a feature may also be used to accelerate further feature processing by being used as an indicator of a pre-established outcome that occurs upon satisfaction or failure of a conditional statement or of structural elements. For example, an indicator may indicate structural elements such as a condition of a vertex being associated with one entity and one score threshold, an outcome of the vertex being associated with two entities and two transaction scores, an outcome of the vertex causing an anomalous program exit for all entities, or the like). By indicating features that most affect a directed graph, indicating similarities between different graphs, or otherwise indicating high-effect parameters and vertices, some embodiments may provide a tool that can reduce issues that are encountered when dealing with a high dimensional parameter space. Such issue reduction may in turn increase efficiency and accuracy when generating, updating, or simulating a program that is modeled by directed graph.
In some embodiments, the present techniques may be implemented using data visualization packages implemented in code such as Python code, Javascript code, or the like. In some embodiments, code implementing the present techniques may be implemented using one or more statistical programming languages such as R and data visualization packages for R, such as Ggplot2. Some embodiments may use Python and data visualization libraries for Python, such as Matplotlib or Seaborn. Some embodiments may use data visualization tools or platforms, such as Gephi™, Neo4j™, Tableau™, Wolfram Mathematica™, Matlab™, or the like. Some embodiments may use web-compatible tools to display data, such as the Javascript library D3js, and permit the initiation or modification of one or more operations described in this disclosure via a user interface (UI) sent over the Internet. Some embodiments may integrate different data visualization tools. For example, some embodiments may implement one or more operations described in this disclosure using the Python Bokeh library, which may allow Python to use one or more elements of Tableau™ to visualized data. Some embodiments may perform just in time compilation of program code and may parse the code to an abstract syntax tree, which may then be transformed into a bytecode representation that is then compiled into machine code (e.g., native machine code of the computer executing the text viewing application, or machine code of a virtual machine).
As shown by the “DocID” property shown in
The arrows 3223 and 3225 may represent a first and second directed edge, each of which is directed from the first vertex 3222 to the second vertex 3224 or the third vertex 3226. The arrow 3223 representing a first directed edge may indicate that the second vertex 3224 may be activated (e.g., set as triggerable) by the satisfaction of the conditional statement associated with the first vertex 3222. Alternatively, or in addition, the arrow 3225 representing a second directed edge may indicate that the third vertex 3226 may be activated by the failure of the conditional statement associated with the first vertex 3222. The relationship between the vertices may be indicated by the “Consequence” property shown for the first vertex 3222, which has a list of sublists as a value, where each sublist includes a vertex, a category label associated with the vertex.
In addition, the fourth vertex 3228 may represent a prohibition vertex, and may include or otherwise be associated with a conditional statement indicating that a first entity “Ent1” is prohibited from satisfying the conditional statement having the identifier “13.” As indicated by the “consequence” property of the fourth vertex 3228, a failure to satisfy the conditional statement indicating that the first entity “Ent1” may cause the activation of the vertex “Vert.95”. While not pictured, a visualization of a directed graph may include a directed edge to a vertex having the vertex identifier “Vert.95” of the type “right.”
As further discussed below, values that are stored in a first data structure representing the first vertex may be used as features for the first vertex. Alternatively, or in addition, values stored in other data structures associated with the first vertex may be used as features for the first vertex. For example, the category label “obligation” of the second vertex 3224 may be used as a feature titled “consequent satisfaction vertex category” of the first vertex 3222.
The directed graph portions enclosed by the box 3220 may be converted into a set of feature values shown in table 3240. In some embodiments, one or more of the features surrounded by the box 3241 may be obtained from a list of pre-determined features. Some embodiments may be able to refer to multiple lists of pre-determine features and choose one or more of the multiple lists based on properties associated with the directed graph of a smart contract program or other elements of the smart contract program. For example, some embodiments may determine that a directed graph is associated with a metadata category “loan” and, in response, indicate that transaction scores associated with a unit of currency should be collected.
Alternatively, or in addition, some embodiments may adapt to different sets of features for specific vertices or vertex types (e.g., a vertex type defined by being a vertex associated with a specific category label) by determining one or more features based on a set of rules, decision trees, or implementations of other symbolic AI operations. For example, some embodiments may determine that one or more vertices surrounded by the box 3220 includes an expiry value and, in response, determine that “expiry” is a feature of the directed graph. Alternatively, or in addition, some embodiments may use a rule or other symbolic AI operation to determine which feature values to collect based on another feature of a vertex, such as a category label. For example, some embodiments may determine that a directed graph having the directed graph portion surrounded by the box 3220 has the category label “obligation” and, in response, obtain features from data that is associated with the vertex and indicated as being present in “obligation” vertices. Such features may include a “conditions,” feature, where a corresponding feature value may include an array of conditional statement identifiers, a “condition threshold” feature, where a corresponding feature value may include one or threshold values or ranges, or the like. Some embodiments may increase visualization and analytical capabilities by including one or more operations to adapt to different vertex data.
Some embodiments may also use one or more NLP outputs to determine a feature value for a vertex. As indicated by the feature values surrounded by the box 3244 and the box 3246, some embodiments may associate a respective vertex with references to a natural language document identifier such as “doc.txt” and a respective text position in the natural language document indicating a respective text section corresponding to the respective vertex. Some embodiments may also include one or more results from an NLP operation (e.g., topics, word frequency, sentiment, or the like) performed on a text section corresponding to the vertex. For example, some embodiments may perform a set of NLP operations to determine a set of embedding values based on the text. Some embodiments may then determine a topic and a topic score based on the set of embedding values. This topic and topic score may then be included as a feature of a vertex, as indicated by the column 3248.
In some embodiments, a category label may be used by some embodiments to more efficiently determine one or more features to prioritize when performing feature selection, feature extraction operations, or other operations to determine a prioritized vertex for modification or display. For example, the category label “obligation,” which is associated with first vertex 3222 and the second vertex 3224, may be used as a reference by some embodiments when performing feature selection or feature extraction. In some embodiments, each category label may be associated with its own set of weights to update feature extraction input values corresponding to a corresponding set of feature values. For example, if the category label “obligation” is associated with the feature “expiry,” some embodiments may assign greater weight to the “expiry” feature of a vertex when performing feature selection operations or feature extraction operations. Alternatively, or in addition, some embodiments may de-prioritize one or more features in response to their association with a category label. For example, in response to a determination that a vertex is categorized as a “right,” some embodiments may reduce a weight associated with the “expiry” feature or eliminate the “expiry” feature from consideration. Some embodiments may increase the accuracy and efficiency of feature or vertex prioritization operations described in this disclosure by modifying the weights of features or eliminating a feature from consideration.
Alternatively, or in addition, a set of prioritization parameters may include a value indicating a visualization type, a vertex indication type, a graph edge type, or the like. For example, some embodiments may indicate a first visualization type, where graph visualizations of the first visualization type include all of the triggered vertices, active vertices, or vertices directly associated with an active vertex via a set of graph edges. Alternatively, or in addition, some embodiments may indicate a second visualization type, where a graph visualization of the second visualization type includes untriggered vertices of a directed graph. In some embodiments, a vertex indication type may indicate a color, a vertex shape, a pattern, a vertex animation, or the like. For example, some embodiments may obtain a prioritization parameter indicating that a prioritized subset of vertices may be colored red, while an un-prioritized subset of vertices may be colored blue.
Some embodiments may obtain the set of prioritization parameters from a graphical UI. For example, some embodiments may visualize a directed graph in a first region of a display of a graphic UI based on parameters obtained from values selected at a second region of a display of the graphic UI. Alternatively, or in addition, some embodiments may use a default set of prioritization parameters, such that even without a direct selection of any prioritization parameters for a specific directed graph, the specific directed graph may be visualized using the default set prioritization parameters. Alternatively, or in addition, some embodiments may obtain the set of prioritization parameters via a message received at an API of a platform executing operations of the process 3300.
In some embodiments, the process 3300 may include obtaining a set of features for vertices of a set of directed graphs, as indicated by block 3304. A set of features may include a set of properties associated with a vertex of the directed graph. For example, the set of features may include a vertex status indicating whether a vertex is satisfied, untriggered, failed, or the like. As discussed above, a set of features for a vertex may include features indicating relationships between different vertices of a directed graph. In some embodiments, a feature for a first vertex may include an indication of a vertex type that the first vertex is associated with via an edge of the directed graph. For example, a feature for a first vertex may be entitled “failure consequence vertex type,” where values for the feature may be selected from a set of categories such as “terminate program,” “penalty condition,” “the payment acceleration condition,” in some embodiments, a feature for a first vertex may include an indication that the first vertex is in a specified graph portion template. For example, some embodiments may determine that a feature or a first vertex may be selected from identifiers of a library of graph portion templates.
A set of features may include a category associated with the vertex or its conditional statement, such as “obligation,” “prohibition,” or “right,” where the categories may be of a category type that restricts the category label to one of a set of mutually exclusive categories. As further discussed below, some embodiments may reduce the set of features for consideration based on a category assigned to a vertex. In some embodiments, a category may be associated with a specific structure of a conditional statement. For example, a vertex or its associated conditional statement may be categorized as a “prohibition.” In response, some embodiments may proceed to extract feature values or related data specific to a prohibition norm vertex. By taking advantage of a specific structure, some embodiments may increase the efficiency and accuracy of feature collection. For example, some embodiments may determine that a vertex is categorized as a prohibition and, in response, apply operations that obtain a first and second element from a consequent vertex (e.g., a consequent vertex identifier and a category label) instead of iterating through a list of unknown length for each vertex in a list of consequent vertices.
Feature values for a set of features for a vertex may include data associated with an entity indicated by the vertex or associated with the corresponding directed graph. For example, the set of features for a vertex may include a first feature “triggering entity,” which may indicate the entity that is directly responsible for satisfying or failing a conditional statement of the vertex. The set of features may also include one or more categories associated with the entity, where the category may include an entity role indicating roles such as resource allocator, resource receiver, third-party agent, or the like.
In some embodiments, the set of features may be based on a pre-determined set of features that are provided as a default set of features. A pre-determined set of features may then be updated via a UI, API-received message, or other computer-interpretable system. Alternatively, or in addition, some embodiments may obtain a set of features using rules or other symbolic AI methods. Alternatively, or in addition, some embodiments may obtain a larger set of features and then filter the set of features using symbolic AI methods during a subsequent operation, such as operations for block 3308.
In some embodiments, the process 3300 may include obtaining a set of feature values of the set of directed graphs, as indicated by block 3306. Some embodiments may obtain the set of feature values or the set of features discussed above from an in-execution application via its corresponding program state or execution environment. Alternatively, or in addition, some embodiments may obtain feature values based on simulations of a directed graph as it is expanded through multiple possible scenarios using one or more simulation operations, as discussed in this disclosure. For example, some embodiments may obtain feature values for each vertex of the directed graph based on data stored in a distributed data storage system such as data stored in an IPFS data storage system or persistent storage of a centralized computing system. Alternatively, or in addition, some embodiments may obtain feature values from data for a completed program using a centralized repository or distributed ledger. For example, some embodiments may obtain feature values of a directed graph as it evolves over a state history or may use a set of numbers, set of strings, set of categories, list of values or the like representing a total directed graph state and usable as a feature value.
As discussed above, some embodiments may obtain feature values encoded directly in a data structure representing a vertex (e.g., a table, a tree, an object, or the like) or a data structure directly associated with a vertex. For example, some embodiments may obtain a vertex category label from an associative array used to encode or otherwise store data of a directed graph. Alternatively, or in addition, some embodiments may perform additional operations on data associated with a vertex to obtain feature values. For example, as discussed above, some embodiments may analyze text sections associated with a vertex and use the analysis results as features of the vertex. Such analysis results may include word frequency counts, sentiment analysis, semantic associations, other NLP operations, or the like.
Some embodiments may include data associating neighboring vertices as a feature value of a directed graph. Some embodiments may use a transaction score associated with a conditional statement of each child vertex of a first vertex of a directed graph as feature values of the directed graph or the first vertex. For example, some embodiments may include a predicted score, such as an anticipated reward value. The predicted score may be based on outcomes of child vertices and their likelihood of occurrence using one or more operations associated with a prediction operation disclosed in this disclosure and may be determined using one or more neural networks disclosed in this disclosure. For example, some embodiments may associate a reward value determined using a trained neural network such as one of the trained neural networks described above. As further discussed below, by incorporating data determined based on child vertices, some embodiments may incorporate predictive modeling results into one or more visualization operations or feature selection operations described in this disclosure. Additionally, some embodiments may increase, decrease, or otherwise modify a feature value in response to a determination that a first vertex is associated with a first category label and that a first conditional statement associated with the first vertex is satisfied. For example, some embodiments may use a set of features that includes a feature titled “consistency.” In response to a determination that a first vertex is associated with the category label “obligation” and a satisfaction state “satisfied,” some embodiments may increase the feature value for the feature titled, “consistency.” Similarly, in response to a determination that a second vertex is associated with the category label “prohibition” and a satisfaction state “failed” (e.g. a condition triggering the second vertex is satisfied), some embodiments may decrease the feature value for the feature titled, “consistency.”
In some embodiments, the process 3300 may include determining a set of updatable features based on the set of features and one or more parameters of the set of prioritization parameters, as indicated by block 3308. As discussed above, features may include various data associated with a directed graph, conditional statements associated with the directed graph, data of an application modifying a directed graph stored in program state, or the like. One or more parameters of the set of prioritization parameters may include a set of update criteria that indicate features that may be updated. An updatable feature may be contrasted with a feature that is not updatable based on the set of update criteria. For example, the duration for which an amount of computing memory is allocated may be indicated as updatable by a first update criteria for a first program, whereas the duration for which the amount of computing memory is allocated may be indicated as not updatable by a second update criteria for a second program. In some embodiments, determining a set of updatable features may increase the domain-specific applicability of simulations and visualizations of directed graphs representing smart contracts. In addition, a set of update criteria may include ranges in which a feature may vary, which may be used when prioritizing features, as further discussed below.
In some embodiments, the set of update criteria may be based on a pre-determined set of features. For example, the update criteria may be set based on a list of features ‘[“total amount”, “resource type”, “expiry”, “satisfied consequence reward value”, “failed consequence reward value”]. Alternatively, or in addition, some embodiments may determine the set of update criteria based on a comparison of values across multiple smart contracts or other symbolic AI programs. For example, some embodiments may determine that a first feature “xcs102—failure penalty” is an updatable feature based on the feature value for this feature varying across a plurality of smart contracts, where this feature may represent an amount that is penalized for failing a condition associated with the conditional statement identified by the identifier “xcs102.” Some embodiments may determine that a second feature “time before termination” is not an updatable feature based on the values for this feature remaining the same across the plurality of smart contracts. By determining the update criteria for a directed graph, its associated program, or applications that shared similar properties as the associated program, some embodiments may detect or indicate relevant features across a variety of different environments.
In some embodiments, the process 3300 may include performing a feature selection operation based on the set of updatable features or the set of prioritization parameters to obtain a prioritized subset of features, as indicated by block 3316. A feature selection operation may include implementing one or more feature selection methods to determine a subset of features to focus on when generating or otherwise updating a visualization of a distributed smart contract program or other symbolic AI program. A feature selection operation may determine a plurality of candidate subsets of features and, for each respective candidate subset, determine a respective score associated with the respective candidate subset of features. Some embodiments may then select one or more of the subsets of features as features to indicate in a UI or other visualization. The feature selection operation may be associated with a pre-determined outcome metric such as an amount of a resource to be allocated, an amount of a resource to be transferred, an amount of a resource to be received, a net amount of a resource change over a period of time, or the like. For example, some embodiments may determine outcome scores based on a net change in the amount of a digital currency stored by an entity after a five year period. Various feature selection methods may be used to select features of the vertices of a directed graph, such as feature selection methods listed in Kumar et al. (Kumar, V. and Minz, S., 2014. Feature selection: a literature review. SmartCR, 4(3), pp. 211-229) or Deng et al. (Deng, X., Li, Y., Weng, J. and Zhang, J., 2019. Feature selection for text classification: A review. Multimedia Tools and Applications, 78(3), pp. 3797-3816), both of which are hereby incorporated by reference.
Some embodiments may use wrapper methods to determine a subset of features, such as those described by Kohavi et al. (R. Kohavi and G. John, “Wrappers for feature subset selection”, Artificial intelligence 97.1-2 (1997): 273-324), which is hereby incorporated by reference. A wrapper method may include computing models with different subsets of features and iteratively determining an optimal subset of features. One or more various types of wrapper methods may be implemented, such as a forward selection method, a backward selection method, a stepwise selection method, or the like. For example, some embodiments may use a forward selection method that includes determining a feature subset score for different features of a set of features. A set of feature subset scores may include a statistical measure, such as a statistical p-value of a statistical T-test or statistical F-test, and may be used to determine a subset of features. Alternatively, the feature subset score may be computed using other functions, such as functions that determine a range, variance, covariance, or the like. Using the forward selection method may include updating a subset of features with additional features from the set of features and varying their corresponding feature values until the corresponding feature subset scores stop improving or otherwise satisfies a feature subset score threshold. For example, some embodiments may add features to a subset of features and compute a corresponding p-value until a minimum p-value is reached. Alternatively, or in addition, some embodiments may use a backward elimination method that includes determining a statistical output from the full number of features and then eliminating a feature from consideration based on which features may be removed with minimal impact on an output value. Alternatively, or in addition, some embodiments may use a recursive selection method when implementing a wrapper method, which may include operations to iteratively generate subsets of features until a computed feature subset score no longer improves (e.g., has reached a maximum value or minimum value based on value).
Some embodiments may use a filter method to determine a subset of features, where a filter method may include determining statistical measures as scores for different subsets of features. Some embodiments may determine the statistical measure with filter methods such as an analysis of variance (ANOVA) method, a Pearson correlation, variance thresholding, or the like. For example, some embodiments may implement an ANOVA method to determine variations of an outcome score based on changes in different features of the updatable features. Using the ANOVA method may include determining an F statistic for each individual feature, where the F statistic may be based on a ratio of the variation between a correlation between a feature subset score and feature values in the numerator and the variation between a correlation between a feature subset score and feature values within feature values of the denominator. The test statistic may then be tested against a threshold and, if determined to have satisfied the threshold, may cause some embodiments to include the feature in a prioritized subset of features.
Some embodiments may also use a combination of filter methods. For example, some embodiments may first use an ANOVA method to determine a first subset of features from the set of updatable features and then use a variance thresholding method to determine a second prioritized subset of features. Using the variance thresholding method may include determining the variance of a feature across a plurality of feature values and associating a predicted power of the feature with its variance. For example, a set of vertices may each be associated with a conditional statement encoding an obligation of a first entity to allocate a varying amount of memory to a second entity, where each vertex may be associated with a predicted reward value. Some embodiments may then determine the variance of a feature titled “allocated memory amount” based the predicted reward values and compare the variance to a variance threshold, where exceeding or otherwise satisfying the variance threshold may cause some embodiments to include the feature “allocated memory amount” in a prioritized subset of features.
In some embodiments, the feature selection operation may include implementations of embedded methods. An embedded method may include operations to determine prioritized features using a trained machine learning model. After the machine learning model of an embedded method is trained, some embodiments may use the trained model to determine feature importance. For example, some embodiments may use a regularization model when training a machine learning model for feature selection operations, where the regularization model may include a Lasso regression (L1) model, a ridge regression (L2) model, an L1/L2 regularization model, or the like. For example, some embodiments may implement an L1 model, where some embodiments may use a trained model (e.g., a trained neural network) to determine a cost function as a score while applying a penalty term to the cost function that reduces the coefficients associated with features and may be tuned to remove features from a set of updatable features.
Alternatively, or in addition, some embodiments may implement an embedded method that uses a tree method to determine scores for each feature and select a most relevant set of features. A tree method may include a regression tree, a classification tree method, a random forest method, or the like. For example, some embodiments may use a tree-based embedded method by using a trained random forest to predict an impurity value as a feature subset score, where each feature of a set of features may correspond with a node of the random forest. Some embodiments may then determine a prioritized subset of features based on a maximum or minimum feature subset score. For example, some embodiments may determine a prioritized subset of features based on which subset of features was computed using a trained tree model to produce the greatest impurity value.
In some embodiments, the process 3300 may include determining a set of extracted feature values based on the set of updatable features and the prioritization parameters, as indicated by block 3320. A value of an extracted feature may be generated using a set of feature extraction input values that may include a set of existing feature values and may be computed from statistical operations, machine learning operations, or the like. For example, a set of vertex features may include a transaction score, expiry duration, assigned probability of triggering, and some embodiments may generate an extracted feature based on a linear or nonlinear combination of these three features. In some embodiments, an extracted feature value may be determined for an entire directed graph or a directed graph portion. Alternatively, or in addition, extracted features may be used to determine similarities between a vertex and other vertices, which may then be used to determine which vertices to modify or prioritize in a graphical display.
Some embodiments may use a feature extraction parameter to determine an extracted feature value. A feature extraction parameter may include a value used to determine an output of a feature extraction operation. For example, a feature extraction operation may include using a neural network to determine one or more results, where a feature extraction parameter for a neural network may include a weight or bias of a neuron of the neural network, a number of layers of the neural network, another hyperparameter of the neural network, or the like. A feature extraction parameter may also include an extracted feature threshold. In some embodiments, a vertex having an extracted feature that satisfies a corresponding extracted feature threshold may be associated with a prioritized vertex or non-prioritized vertex, as further discussed below. In some embodiments, satisfying an extracted feature threshold may include having a corresponding extracted feature value satisfying the extracted feature threshold. For example, some embodiments may obtain features of a first vertex to compute an extracted feature value for the first vertex using a trained neural network, where the output of the neural network is equal to 0.65, which may satisfy an extract feature threshold equal to 0.5.
Some embodiments may perform statistical operations based on a set of feature values associated with a directed graph, such as the set of feature values determined above. Performing a statistical operation may include implementing one or more statistical methods, such as analysis of variance (ANOVA), ANOVA-simultaneous component analysis (ANOVA-SCA), PCA, or the like. Some embodiments may include a library of extracted feature values that includes the set of features to use for the statistical operation, a threshold(s) or range(s) for values of the extracted features, or indicators of whether values satisfying the threshold(s) or within the range should be prioritized based on prioritization parameters. For example, some embodiments may retrieve a set of eigenvectors (sometimes known as “principal component vectors” in the context of PCA operations) from a database or otherwise obtain the set of eigenvectors based on a category or other value associated with a directed graph (or its corresponding program). Some embodiments may then obtain a first set of extracted features by determining a set of coefficients represented as a multidimensional vector. In some embodiments, adding a linear combination of the eigenvectors weighted by the corresponding elements of the multidimensional vector may provide an output value equal to or otherwise similar to a set of feature values for a directed graph or directed graph portion, where similarity between vectors may be include all same-index elements of the vectors being within 5% of each other (e.g., the first element of a first vector is within a 5% range of the first element of a second vector). While the above uses 5% as a similarity threshold, over values may be used, such as values less than 20%, less than 10%, less than 5%, less than 1%, or the like.
Some embodiments may use one or more unsupervised learning operations to determine a set of extracted feature values for a directed graph, a portion of the directed graph, or a vertex of the directed graph. For example, some embodiments may use clustering to determine an extracted feature value. Features such as a net transaction score, a cumulative predicted reward value, a total text sentiment, or a category label may be used to determine a first position for a directed graph in a first feature space (e.g., a vector space, where each dimension of the vector represents a feature). Similarly, features such as a transaction score, a predicted reward value, text sentiment, or a category label may be used to determine a first position for a vertex in a second feature space. Some embodiments may calculate a distance in feature space from the first position to a cluster centroid and associate the vertex to the cluster centroid. Some embodiments may then determine an extracted feature value based on a library of clusters or other historical values associating specific prioritization categories with clusters or cluster centroid based on the feature, where clusters or cluster centroids may be associated with a category.
Some embodiments may determine an extracted feature value using one or more neural networks or other machine learning systems, such as those described in this disclosure. One or more types of neural networks may be used to generate a vector representing a directed graph, which may then be used to determine similarities between directed graphs. Similarly, one or more types of neural networks may be used to determine whether a vertex should be included in a prioritized subset of vertices. For example, some embodiments may use an autoencoder to determine an extracted feature score for a directed graph based on feature values of the graph, such as an autoencoder described by Dong et al. (Dong, G., Liao, G., Liu, H. and Kuang, G., 2018. A review of the autoencoder and its variants: A comparative perspective from target recognition in synthetic-aperture radar images. IEEE Geoscience and Remote Sensing Magazine, 6(3), pp. 44-68). Using an autoencoder may include using a neural network having a same number of output layers as input layers, such as using a regularized autoencoder, a variational autoencoder, or the like.
Some embodiments may use a neural network adapted to or adaptable for analyzing a graph for the purposes of classifying whether one or more vertices should be prioritized. For example, some embodiments may use a graph neural network (GNN) such as one described by Zhou et al. (Zhou, J., Cui, G., Zhang, Z., Yang, C., Liu, Z., Wang, L., Li, C. and Sun, M., 2018. Graph neural networks: A review of methods and applications. arXiv preprint arXiv:1812.08434) or Zhang et al. (Zhang, S., Tong, H., Xu, J. and Maciejewski, R., 2019. Graph convolutional networks: a comprehensive review. Computational Social Networks, 6(1), p. 11), both of which are hereby incorporated by reference. A GNN may include an operation to determine a set of adjacency values between different vertices of a directed graph before or after using one or more layers of a neural network to determine an output value of the GNN. For example, some embodiments may use a GNN to classify a directed graph (or a portion of the directed graph). Using the GNN to classify the direct ed graph may include a determining a product (e.g., a matrix multiplication product) of a set of feature values for the directed and a vector based on an adjacency list (or other data structure encoding vertex adjacency) representing the set of directed edges connecting vertices of the directed graph. A graph convolutional neural network (or another neural network) may be trained to classify directed graphs or directed graph portions based on various types of priorities, where parameters of the neural network may be included in the set of feature extraction parameters.
Some embodiments may modify the set of feature values used for feature extraction operations by applying one or more weights to a set of updatable features, such as the set of updatable features determined using operations described above for block 3308. Alternatively, or in addition, some embodiments may include operations to use the prioritized subset of features determined using operations described for block 3316 when determining one or more extracted feature values. For example, some embodiments may change a weight associated with each feature of the prioritized subsets when determining extracted features. Some embodiments may, in response to instructions to filter out vertices that already likely to be prioritized, reduce a weight of a prioritized vertex. Alternatively, or in addition, some embodiments may increase the weight associated with each feature of the prioritized subset of features.
As discussed above, some embodiments may include feature values determined from simulation results. For example, some embodiments may determine a feature value for a vertex based on a measure of centralized tendency (e.g., a mean average) of an outcome score associated with 100 simulations, where a starting point of the simulation indicates that the first vertex is active. By executing one or more operations discussed in this disclosure in a simulated environment or obtaining feature data based on simulated information, some embodiments may be used to prepare or re-construct a smart contract program to better meet a target metric (e.g., reduced computing use, increase a transaction score, or the like).
In some embodiments, the process 3300 may include determining a prioritized subset of vertices based on the set of extracted feature values or prioritized subset of features, as indicated by block 3324. As discussed further below, a prioritized subset of vertices may be visually indicated on a graphical display or otherwise prioritized for parameter modification. The prioritized subset of vertices may be indicated by altering a property or value associated with each vertex of the prioritized subset of vertices. Alternatively, or in addition, some embodiments may update a list of vertices referencing each respective vertex of the prioritized subset of vertices.
As discussed above, some embodiments may include one or more vertices in a prioritized subset of vertices based on the feature values of the vertex satisfying a set of prioritization thresholds. The set of prioritization parameter disclosed for block 3302 may include one or more thresholds of the set of prioritization thresholds. For example, some embodiments may include a first vertex in a prioritized subset of vertices based on a feature value of a prioritized feature satisfying a prioritization threshold, where satisfying the prioritization threshold may include allocating more than a threshold number of digital assets.
Alternatively, or in addition, some embodiments may determine a vertex to be a prioritized vertex based on an extracted feature value of the vertex satisfying one or more prioritization thresholds. For example, some embodiments may determine that a vertex is a prioritized vertex based on the vertex embedding value determined from an autoencoder satisfying an embedding threshold. Alternatively, or in addition, some embodiments may determine a vertex to be a prioritized vertex based on a combination of feature values satisfying a combined feature value threshold. For example, some embodiments may determine that a total number of score changes (e.g., a net change in a computing resource amount, a net change in a stored number of a digital asset type, a total elapsed time from a starting point, or the like) associated with a set of vertices satisfies a threshold. In response, some embodiments indicate all of the vertices or the N vertices having the greatest effect, where N may be one of various integer values (e.g., N=2, 3, 4, 5, a number greater than 5).
In some embodiments, the prioritization parameters may include a parameter indicating that a vertex causes or is otherwise associated with an anomalous program exit. Some embodiments may exit a smart contract program by terminating an execution of the smart contract program on a computing platform. A smart contract program may be exited via an authorized outcome. For example, after receiving an indication of a full repayment of an outstanding balance, some embodiments may determine that an entity has no further obligations and end an execution of the smart contract program. Alternatively, or in addition, a smart contract program may be exited via an anomalous program exit, where the anomalous program exit may be pre-categorized as such or be determined to be an anomalous program exit based on one or more anomalous exit criteria being satisfied. For example, some embodiments may determine that a prohibition vertex is associated with an anomalous program exit for a program based on an outcome of satisfying a conditional statement of the prohibition vertex causing the termination of the program without one or more outstanding obligations being fulfilled.
In some embodiments, the process 3300 may include determining a previously-recorded graph portion, has indicated for block 3330. A previously-recorded graph portion may be a graph portion of a directed graph that shares one or more similar features with another graph portion of the same directed graph or a graph portion template. For example, some embodiments may determine that a first graph portion includes a first vertex labeled as an “obligation”, a directed edge associating the first vertex to a second vertex labeled as “prohibition,” the second vertex, and a directed edge associating the second vertex to a third vertex labeled as “prohibition.” Some embodiments may then determine that the first graph portion matches a graph portion template (e.g., a graph portion template stored in a library of graph portion templates) and, in response, indicate the graph portion as a previously-recorded graph portion.
In some embodiments, the process 3300 may include determine cross-graph metrics based on the set of extracted feature values, as indicated by block 3334. As discussed above, some embodiments may include compressed information representing one or more directed graphs of a set of directed graphs. Some embodiments may determine a similarity score between different graphs to determine similarity values between directed graphs, where the similarity scores may be calculated based on a function (e.g., a difference measurement, a distance measurement, or the like) based on an extracted feature value. In some embodiments, the extracted feature value may include a value determined using operations described for block 3320. For example, a first directed graph may be reduced to a first vector titled “graph-vec1” in a 100-dimension vector space using a GNN and a second directed graph may be reduced to a second vector titled “graph-vec2” in the 100-dimension vector using the GNN. Some embodiments may then determine that the two directed graphs are similar based on satisfying one or more vector space distance thresholds (e.g., a Euclidean, cosine, Minkowski, or Manhattan distance between the first vector and second vector are less than a vector space distance threshold) and, in response, determine that the two directed graphs are similar to each other.
Some embodiments may modify a similarity score based on one or more vertex of a prioritized subset of vertices or one or more features of a prioritized subset of features, such as by increasing their effective importance with a set of weights. For example, features of a prioritized subset of vertices determined using one or more operations of block 3324 may be weighted two-fold to increase their effect during a PCA analysis. In some embodiments the importance of certain features may be weighted more heavily when determining a similarity score, as described above. Some embodiments may apply a set of weights that increase the effect of a specific category label, such as a category label selected from a set of mutually exclusive category labels. For example, some embodiments may set each category label “right,” “obligation,” and “prohibition” to be its own feature when determining features and feature values for block 3304 and block 3306, and then weighting (e.g., multiplying by) each category label by a value greater than one relative to other feature values (e.g., multiply by a number greater than one, greater than two, greater than five, or the like). Alternatively, or in addition, some embodiments may decrease the effect of a prioritized subset of vertices or features, such as by multiplying feature values by weights that are less than one.
Some embodiments may increase the likelihood of two directed graphs satisfying a similarity threshold based on similar graph portions. For example, some embodiments may determine that two directed graphs having graph portions that match with a same graph portion template. In response, some embodiments may increase a similarity threshold, where a determination that a Euclidean distance value between vectors representing the two directed graphs is less than the similarity threshold causes some embodiments to determine that the two directed graphs are similar. Some embodiments may use a similarity metric or result determined during the computation of the similarity metric to determine vertices or graph portions that are similar to each other. The similarity metric may be determined based on a measure of distance between vectors in a feature space. For example, the similarity metric may include a Euclidean, cosine, Minkowski, or Manhattan distance between a first vector and a second vector.
In some embodiments, the process 3300 may include indicating the set of cross-program similarity metrics, the prioritized subset of vertices, or the prioritized subset of features, as indicated for block 3344. Indicating the metrics, vertices, or features may include displaying a visualization of a directed graph, where the visualization of the directed graph may include shapes, text, or other visual features on a computer screen or three-dimensional (3D) projection. For example, some embodiments may display a visualization of a directed graph that uses two-dimensional (2D) shapes that include triangular, rectangular, circular, or the like as indicators for vertices of a directed graph.
Some embodiments may display set of similar directed graphs based on the set of similarity metrics. For example, some embodiments may provide a list of directed graphs similar to a first directed graph, where each directed graph of the list of directed graphs is determined to be similar to the first directed graph using one or more operations described for block 3334. Alternatively, or in addition, some embodiments may display a second graph representing relationship between different directed graphs in a multidimensional vector space.
Some embodiments may visually distinguish a prioritized vertex by showing a corresponding vertex visualization as having a different color, a different size, a different shape, a different animation (e.g., vibrating), or the like. Some embodiments may include a UI element that causes a visualization region to visually distinguish the prioritized vertex, where the UI element may include a button, slider, menu, or the like. For example, some embodiments may include a visualization region that was initially displaying 50 vertices of a directed graph and an arrow button. Clicking on or pressing on the arrow button may cause the visualization region to focus on a graph section showing the prioritized vertex and highlight the prioritized vertex in a different color or cause the prioritized vertex to animate (e.g., vibrate, expand or contract in size, or the like) or otherwise be distinguishable from one or more surrounding vertices.
Some embodiments may display a directed graph in the form of a charged-particle visualization or other visualization emulating a physical system. The charged-particle visualization may include 2D or 3D shapes as representations of vertices and lines between the 2D or 3D shapes as representations of directed edges. As used in this disclosure, a particle of a charged-particle visualization be shaped as or otherwise have an outline in a shape such as a circle, a polygon, a sphere, a polyhedron, or the like. Some embodiments may use quantitative values to determine a distance between two particles in a visualization, where quantitative values may include a likelihood of occurrence, a predicted reward value, or the like.
Some embodiments may determine or update a size, a color, or other property of a vertex visualization based on a value associated with the vertices. The value may be used to determine a property of the vertex visualization and may include a transaction score associated with a conditional statement of the vertex, a consequence score associated with failing the conditional statement, a consequence score associated with a satisfied conditional statement, or the like. For example, some embodiments may update a size of a particle representing a vertex based on a threshold value of a conditional statement of the vertex.
Some embodiments may use associations between sections of a natural language document and vertices or edges to update a visualization of a directed graph. For example, some embodiments may display a prioritized vertex and a sidebar, the sidebar including links to or text from a natural language document. The links or text sections may be associated with the prioritized vertex based on a text section reference value included in a set of properties associated with the prioritized vertex. For example, a first vertex may include or otherwise be associated with the array [“sample text.doctype”, 1562, 1583],” which may be a reference to a first text section starting at character position 1562 and ending at character position 1583 of the natural language document “sample text.doctype.” When visualizing a directed graph that includes the first vertex, some embodiments may display or otherwise indicate text from the first text section.
Alternatively, some embodiments may show a directed graph that includes only triggerable vertices, triggered vertices, or child vertices of the triggerable vertices. For example, some embodiments may show the directed graph includes the obligation vertex 3430 and is not include the first vertex 3432. Some embodiments may be capable of providing visualizations that do include and do not include un-triggered vertices, where one or more UI elements may be toggled show or not show the un-triggered vertices.
The vertices 3412-3414 of the directed graph may be a first prioritized subset of vertices determined using one or more of the operations described in this disclosure. The vertices 3412-3414 may be visually indicated with a different color, box size, or may otherwise visually distinctive from other vertices of the directed graph shown in the first region 3402. For example, as shown in
The reduced graph portion vertex 3480 may be represented as a larger vertex relative to other vertices in the directed graph shown in the first region 3402. The reduced graph portion vertex 3480 may represent a previously-recorded graph portion, such as a previously-recorded graph portion determined using one or more operations described for above for block 3330. For example, the reduced graph portion vertex 3480 may represent a directed graph portion made of three vertices having the identifiers “R4,” “S4,” and “S5,” which may be matched to a graph portion template. Matching the three vertices to a graph portion template may include determining that the graph portion template includes vertex templates having category labels that match one of the three vertices and directed edge templates having directionality that match edges of the directed graph portion. Additionally, while the vertices of the directed graph in the first region 3402 are shown as boxes, other embodiments may show vertices as other shapes such as spheres, circles, two-dimensional polygons, three-dimensional polyhedrons, or the like.
The graph portion 3418 may represent a graph portion that is indicated to be similar to other graph portions, such as the directed graph identifiers “GR01” and “GR86” written in the fourth region 3490, described further below. In some embodiments, the vertices and edges of the graph portion 3418 may be visually distinguished by changing a color of the graph portion, animating the vertices and edges, changing a surface pattern of the graph portion, adding a halo effect to the graph portion, or the like. By indicating graph portions as similar, some embodiments may increase the efficiency of graph outcome predictions or analysis.
The second region 3450 shows a list of features and manipulatable values of the set of features as surrounded by the dashed box 3453. Some embodiments may determine the list of features to indicate based on a prioritized subset of features determined using one or more operations described for block 3316. The text boxes shown in the dashed box 3453 may be used to update a set of feature values corresponding to the set of features on their left. For example, the value 37 corresponding to the obligation vertex 3422, which is identified by “03,” may be changed to 80, which may update a threshold value of the directed graph represented in the first region 3402. Additionally, some embodiments may limit the values used to update a feature value. For example, some embodiments may limit the range of the feature value “expiry” associated with the vertex labeled R3 (i.e., vertex 3414) to range between 1 and 100. The directed graph may be stored in persistent memory or non-persistent memory in a simulated environment.
The third region 3460 may include a set of UI elements that provide options to select, update, or otherwise determine one or more prioritization parameters. The first UI element 3462 may include a drop-down menu to select a first prioritization type, where the first type of privatization value may indicate a category for prioritization. The second UI element 3464 may indicate a specific prioritization parameter, such as “threshold.” The third UI element 3466 may include a data entry box that may display the value of the specific prioritization parameter shown in the second UI element 3464, where the third UI element 3466 may be edited to modify the value for the prioritization parameter shown in the second UI element 3464.
Some embodiments may determine that a prioritization parameter has been updated and, in response, update a prioritized subset of vertices. For example, some embodiments may determine that the third UI element 3466 has been updated, which may then update the prioritization parameter “threshold” of the prioritization category “critical failure.” Some embodiments may then re-determine the prioritized subset of vertices and modify the display of the directed graph shown in the first region 3402. For example, some embodiments may determine a prioritized subset of vertices that does not include the vertex 3414 and display a version of the directed graph shown in the first region 3402 that does not visually distinguish the vertex 3414 from other non-prioritized vertices such as the obligation vertex 3430.
The fourth region 3490 may include a set of UI elements that include identifiers of directed graphs determined to be similar to the directed graph shown in the first region 3402. The fourth region 3490 indicates the label “portion01,” which is represented by the graph portion 3418 shown in the first region 3402. As described above, the graph portion 3418 may include a graph portion determined to be similar to another graph, such as the directed graph identified as “GR01” or the directed graph identified as “GR760.” In some embodiments, the UI elements of the fourth region 3490 may also display similarity metrics indicating the similarity between a directed graph or a directed portion of a first directed graph and a corresponding directed graph or a directed portion of a second directed graph. For example, the UI elements of the fourth region 3490 displays the value “0.21” to indicate that a extracted feature vector of the directed graph “GR01” has a normalized distance equal to 0.21 in a multidimensional vector space from an extracted feature vector of the directed graph portion 3418, where the extracted feature vector may be determined using one or more operations of the process 3300. Similarly, the UI elements of the fourth region 3490 displays the value “0.03” to indicate that an extracted feature vector of the directed graph “GR86” has a normalized distance equal to 0.03 in a multidimensional vector space from the extracted feature vector of the directed graph portion 3418.
described above, some embodiments may use one or more machine learning operations to determine outcome scores, outcome states, or the like. Some embodiments may adaptively transfer parameters from one machine learning model to another to increase the efficiency or accuracy of results obtained from a machine learning model. Some embodiments may perform operations, such as those described further below, to transfer parameters of machine learning models responsive to properties of a directed graph.
Adaptive Parameter Transfer for Learning
Various applications may use machine learning models to detect issues, optimize performance, make predictions, or the like. However, performing the training necessary to optimize learning operations may be computationally costly for end-user devices. Moreover, the bespoke nature of self-executing protocols that programmatically codify agreements between entities may reduce the amount of training data that would be appropriate for determining model parameters. Furthermore, the privacy of a respective self-executing protocol in a set of self-executing protocols may prevent entities that do not have permission to view the respective self-executing protocol from obtaining values stored in the respective self-executing protocol. Such restrictions may exacerbate the difficulty that an entity may have when obtaining data for determining an accurate set of model parameters.
Some embodiments may transfer learning operations to collect features of multiple self-executing protocols or other symbolic artificial intelligence (AI) models. Some embodiments may then train a machine learning model, such as a neural network, on the data using one or more computing devices to determine a set of model parameters for the machine learning model in association with the labels determined from shared properties of directed graphs of the multiple self-executing protocols. Some embodiments may then export the trained model parameter values to other computing devices without leaking the underlying training data to the other computing devices.
As used in this disclosure, a directed graph encoding or portion of a codified agreement may include a set of records representing a log of states or sequence of events causing the state sequence represented by the log of states, where a log of states may be stored as a set of values that change over time. For example, some embodiments may include or have access to a set of records associated with different timepoints, where each respective record of the set of records stores values of a program state evolving over time at a corresponding respective timepoint. As described further below, some embodiments may determine a graph portion template by determining that a change in program state matches a pattern of change in the set of values. For example, some embodiments may train a machine learning model based on a set of records associated with different time points. Some embodiments may determine that a a stored pattern, such as a a graph portion template, matches with a set of records and, in response, associate the trained learning model with the stored pattern. Alternatively, or in addition, some embodiments may use an indicator that the stored pattern is present as a feature when training the machine learning model to determine a set of model parameters. Alternatively, or in addition, some embodiments may transfer the set of model parameters to an entity. Alternatively, or in addition, some embodiments may receive a version of the set of model parameters and perform a second training operation to update the version of the set of model parameters or determine additional model parameters using an error-correcting model parameter.
When sending the model parameter values to other computing devices, some embodiments may then obtain first a directed graph encoding a self-executing protocol from an entity via a message. Some embodiments may then determine a set of features based on the directed graph, where the set of features may include a graph portion template (e.g., arrays or hash maps representing vertices and edges of the graph portion template, an identifier of the graph portion template, or the like). The graph portion template may match with a graph portion of the directed graph via a shared set of categories for the vertices of the graph portion and the graph portion template. Some embodiments may then obtain one or more of the sets of model parameter values described above based on the graph portion template, another feature of the directed, or a set of features of the directed graph. Some embodiments may then export the set of model parameter values to an address associated with the entity that sent the message. Receiving the set of model parameter values may cause a computing device to determine an outcome score based on the set of model parameter values.
By pre-determining parameters for a machine learning model, some embodiments may increase the efficiency of a training operation, classification operation, or other outcome-score-determining operation be performed on a computing device. Additionally, some embodiments may protect the privacy or security of agreements and their corresponding participant entities by separating model parameter users from operations that use the data from the agreements to determine the model parameter values. Furthermore, by basing a machine learning model on a set of features, some embodiments may be adaptive to different types of graphs and select different types of learning models or subtypes of learning models based on specific geometric configurations of a directed graph. Moreover, by including vertex types or graph portions in the set of features used to determine or select a set of model parameters, some embodiments may provide a mechanism that accounts for vertices of a directed graph that, when triggered or active, may have a significant influence on an outcome score. Such mechanisms may be especially useful in contexts such as the planning and execution of codified agreements between different entities, where a change to single feature associated with a single vertex may alter the properties of an agreement even when relationships between vertices are unchanged. It should be understood that, while some embodiments may be described as gaining a specific benefit or performing a specific operation described in this disclosure, not all embodiments described in this disclosure must perform that specific operation or provide that specific benefit, and may perform a set of operations that do not include that specific operation for its own corresponding set of benefits.
Some embodiments may send a message 4112 to a computing system 4120, where the message 4112 may encode the directed graph 4130 in one or more various ways. The message 4112 may include a web message or other application-layer communication. In some embodiments, the message 4112 may include attributes or other values associated with the directed graph 4130, such as a set of entity identifiers for entities participating in the program represented by the directed graph, vertex statuses of the set of vertices of the directed graph, or the like. The message may also include a set of conditional statements of the set of vertices of the directed graph, where a message that includes the set of conditional statements may include identifiers corresponding to the set of conditional statements, a text form of the conditional statement, or the like. The message may also include a set of conditional statement parameters of one or more of the conditional statements. As used in this disclosure, sending a directed graph or a conditional statement may include sending the values stored in program state representing the directed graph or a conditional statement. Alternatively, or in addition, sending the data object may include sending an identifier of the directed graph or a conditional statement stored in program state.
Some embodiments may receive a message at an API of the computer system 4120, where the computer system 4120 may include the peer-to-peer network 4122 and a centralized server 4124. In addition, the computer system 4120 may include or otherwise have access to a cloud computing resource 4150. In some embodiments, one or more peer nodes of the peer-to-peer network 4122 may be assigned to a specific entity, where the specific entity may be able to use or view data store on the assigned peer node. As described in elsewhere disclosure, some embodiments may determine or otherwise update a set of model parameter values for a machine learning model associated with the directed graph 4130. Furthermore, while the computing environment 4100 is depicts as including the peer-to-peer network 4122, the centralized server 4124, and the cloud computing resource 4150, some embodiments may include a computer system that does not include one or more of the aforementioned elements. For example, some embodiments may include a peer-to-peer network without a centralized server.
After receiving the message 4112, some embodiments may then determine a set of features based on the directed graph 4130 and then determine a set of model parameter values based on the set of features. In some embodiments, the set of model parameter values may be stored in the repository 4152 of a cloud computing resource 4150. The set of model parameter values may be stored in a record 4162 of a database 4164, where the database 4164 stored in the repository 4152. While the database 4164 is shown as being stored in the repository 4152 in this example, some embodiments may store a database that includes the set of model parameter values in another location, such as on a tamper-evident, distributed ledger executing on the peer-to-peer network 4122. In some embodiments, in addition to providing storage for data via the repository 4152, the cloud computing resource 4150 may provide a processor or other computing resources to perform one or more operations described in this disclosure.
After obtaining the set of model parameter values from the record 4162, some embodiments may send the model parameter values to the client computing device 4110. The client computing device 4110 may use the set of model parameter values to determine an outcome score in conjunction with other model parameter values. The other model parameter values may be determined by the client computing device 4110 by performing a local training operation that uses the set of model parameter values from the record 4162. The other model parameter values may be values for downstream operations of a machine learning model after a first using the set of model parameter values 4166. Some embodiments may then obtain a new set of model parameter values from the client computing device 4110, where the new set of model parameter values may be evaluated by the computer system 4120. Some embodiments may evaluate the new set of model parameter values by using the new set of model parameter values in a machine learning model for a testing operation. Some embodiments may then update the previous set of model parameter values in response to a determination that the new set of model parameter values provides a higher accuracy or other measure of performance relative to the previous set of model parameter values.
Obtaining the directed graph may include receiving a message as an application-layer communication sent over a network from a program executing on a client device. In some embodiments, the message may be a web message and may be encoded in the hypertext transport language protocol (HTTP, HTTP2) or according to other application-layer protocols. A message can include one or more transmissions, in some cases with intervening responsive messages, like acknowledgments or API responses.
In some embodiments, the directed graph may be generated via a UI. For example, some embodiments may receive a directed graph created in a UI of a self-executing protocol creation tool (“smart contract creator program”). The smart contract creator program may provide a user with a method of creating a smart contract or other symbolic AI program, where the creation may include selecting, updating, or otherwise determining attributes, properties, values, associations, or the like. For example, the smart contract creator program may include a UI that allows a user to select a set of entities, set of entity roles for a self-executing protocol, set of vertices of the self-executing protocol, set of conditional statements, or the like. Alternatively, or in addition, the smart contract creator program may determine parameters of the conditional statements, global parameters to use as state variables upon which one or more conditional statement depends, time limits for conditional statements, categories for vertices of a directed graph encoding one or more values of the self-executing protocol, edges that associate different vertices together, or the like.
The term “UI” may reference either or both of a static interface for a user or one that involves over time. For example, a first UI may change from one state to the next during an animated transition or in response to a user input. References to objects displayed or actions being performed by a UI do not require that the objects be displayed or actions be performed concurrently. For example, a UI may display a first object or perform a first action and then, at a later point in time, display a second object or perform a second action while still being the same UI as that term is used herein. The term “UI” is used broadly and may refer to something presented to a user, such as by a client computing device. The term “UI” may also refer to instructions or data by which a presentation is created at another computing device that is not displaying the UI, such as a remote server. For example, some embodiments may generate a UI at a server by generating the HTML or JSON code and provide the UI to a client computing device without displaying the webpage itself at the server. The client computing device may then render a webpage based on the code.
In some embodiments, the process 4200 may include obtaining an outcome score type based on data associated with the log of states, as indicated by block 4206. As further discussed below, some embodiments may select different learning models or different model parameters based on an outcome score type. In some embodiments, a user may select one or more outcome score types in a UI and send the selected outcome score type in a message. For example, some embodiments may obtain a first outcome score type, where the corresponding first outcome score of the first outcome score type may indicate whether a program modeled by the directed graph is likely to satisfy a condition of a prohibition vertex, where the prohibition vertex may include a. Alternatively, or in addition, some embodiments may obtain a second outcome score type that indicates whether a program modeled by the directed graph is likely to result in an entity triggering a “rights” vertex such that a condition of the “rights” vertex is activated. Alternatively, or in addition, some embodiments may obtain a third outcome score type, where the outcome score having that third type that indicates whether a program modeled by the directed graph is likely to result in a program triggering an external activity, such as a security audit. Alternatively, or in addition, some embodiments may obtain a fourth outcome score type that indicates whether a program modeled by the directed graph is likely to result in malicious behavior, where behaviors observed for an entity may indicate false or malicious activity of the entity.
The data used to determine an outcome score type may include various types of values, such as a direct identifier or value mapped to an outcome score type, categories of vertices of a directed graph, conditional statement parameters, entity-related data, or the like. For example, some embodiments may obtain an outcome score type by selecting the outcome score type based on an entity or entity role. As described in this disclosure, an entity role may include a category for an entity, where the entity role may indicate, for entities having the entity role, a set of actions, obligated activities, available permissions, or the like. For example, some embodiments may obtain a message including a directed graph and an entity role and then select an outcome score type based on the entity being categorized as having the entity role.
Some embodiments may obtain a set of outcome score types by default if no specific outcome score types are obtained in a message or otherwise selected. Alternatively, or in addition, after obtaining a message indicating a first outcome score type, some embodiments may select both a first outcome score type and a second outcome score type based on additional data. Alternatively, as described further below, some embodiments may proceed with operations the process 4200 without selecting any outcome score types, such as in the case when an application is designed to use one learning model without additional parameter modification.
In some embodiments, the process 4200 may include determining a first set of features based on the log of states, as indicated by block 4208. As used in this disclosure a feature may include an attribute such as a categorization, value, an identifier determined from data associated with a directed graph, or the like. The first set of features may include various attributes of a log of states, such as a log of states represented by arrays or other data structures representing a directed graph. In some embodiments, the first set of features may include a boolean or count of vertices having a specific vertex type, where a vertex type may be determined based on a category label of the vertex, a vertex status of the vertex, a conditional statement of the vertex, a conditional statement parameter, an entity associated with a transaction that activates or triggers the vertex, an entity role of the entity, or the like. For example, the first set of features for a directed graph may include an indicator of whether a vertex labeled as a “prohibition” vertex and having a conditional statement parameter greater than “111” is present in the directed graph. In some embodiments, the first set of features may include a set of graph portion templates, where each graph portion template matches with a graph portion of the directed graph. As used in this disclosure, an attribute of a directed graph may include any value stored in an instance of a program storing the directed graph or otherwise directly represented via program state of the program. For example, a distributed application may store “111” as a conditional statement parameter of a vertex of a directed graph in program state and use the value “111” as an attribute of the directed graph. Alternatively, or in addition, the first set of features may also include one or more of the outcome score types determined above.
In some embodiments, the set of features may include one or more values based on a stored pattern. For example, the set of features may include an indicator of a match between a stored pattern and a set of records representing the log of states, a count of matches between the stored pattern and a log of states, or the like. The stored pattern may include pattern parameters usable as criteria to determine whether a set of values or types of values change over time to be set to a specified set of values or range of values. For example, a first stored pattern may include criteria that a first variable of a program state is equal to a first category at a first timepoint and that a second variable of the program state is equal to a second category at a second timepoint after the first timepoint. A pattern may match with a set of records based on the set of records satisfying the criteria of the pattern. For example, a set of records may match with the first stored pattern value described above based on the first variable of a program state being equal to the first category at a timepoint t=0 and a second variable of the program state being equal to the second category at a timepoint t=10. Some embodiments may then use, as a feature, a value indicating that the first stored pattern value matches with a log of states of the smart contract.
In some embodiments, the set of graph portion template or another set of stored patterns may be obtained from a library of graph portion templates or another library of patterns. Some embodiments may select a graph portion template to include in the set of graph portion templates based on the graph portion template matching with a graph portion of the directed graph. For example, the directed graph may include a first graph portion having a first vertex categorized as a “right” and a second vertex categorized as an “obligation,” where the second vertex is directed to the first vertex via a directed edge. Some embodiments may search through a set of graph portion templates to select a graph portion template having a first vertex template matching with the first vertex with respect to the category “right,” a second vertex template matching with the second vertex with respect to the category “obligation,” and an edge template matching the directed edge associating the second vertex to the first vertex by associating the first vertex template with the second vertex template. By selecting graph portion templates from a library of graph portion templates based on matches with the obtained directed graph, some embodiments may obtain weights or values associated with the directed graph templates known to have an effect on an outcome score, as described further below.
Some embodiments may apply a search method to select or determine graph portions with which to match to graph portion templates based on a directed graph. For example, some embodiments may generate a set of graph portions using a DFS operation or BFS operation to generate a set of graph portions and corresponding graph portion templates. Additionally, some embodiments may generate a set of graph portions using a heuristic search method, such as a Best First search method, A* search method, or the like. A graph portion template generated from a corresponding graph portion may include a set of vertices or set of edges similar to the vertices or edges of the graph portion, where one or more associations or attributes of the vertices or labels may be stripped out. For example, some embodiments may generate a graph portion template from a graph portion having a first vertex, a second vertex, and a directed edge directed from the vertex to the second vertex. The generated graph portion template may include a first vertex of the graph template (vertex template), a second vertex template, and a directed edge directed from the first vertex template to the second vertex template (edge template), where the first vertex template may be associated with the same category as the first vertex (e.g., “Right,” “Obligation,” “Prohibition”) or the like. In some embodiments, the generated graph portion template may be generated without any associations with a conditional statement. Alternatively, or in addition, some embodiments may generate a graph portion template where one or more vertices may be associated with a specific conditional statement, a template of conditional statement, parameters of a conditional statement or the like.
In some embodiments, the process 4200 may include selecting a machine learning model based on the first set of features, as indicated by block 4212. As described further below, some embodiments may use a machine learning model to determine an outcome score. Some embodiments may select a machine learning model such as a supervised learning model (e.g., random forest model, support vector machines model, convolutional neural network, recurrent neural network, or the like), semi-supervised learning model, reinforcement learning model, or the like. Some embodiments select one or more learning models based on a set of outcome score types similar to those described for block 4206 or the first set of features described above for block 4208. Furthermore, some embodiments may select one or more learning models based on the first set of features. By selecting a machine learning model based on the first set of features, some embodiments may increase the robustness and accuracy of the machine learning model being used with respect to specific needs or considerations associated with one or more features of the first set of features.
Some embodiments may select a neural network having directed connections between neurons organized in a temporal sequence, such as a recurrent neural network (RNN). Various types of a RNN may be used, such as a long short-term memory (LSTM) neural network, fully recurrent neural network, Elman network, Jordan network, neural network having a gated recurrent unit, Hopfield neural networks, Bi-directional RNN, continuous time RNN, memristive network using memristor materials, or the like. For example, some embodiments may select a LSTM model to determine an outcome score, categorize a program state based on the log of states leading up to the program state, or the like.
As described elsewhere in this disclosure, some embodiments may use a graph neural network to determine an outcome score and classify a directed graph or portion of the directed graph based on the outcome score. The graph neural network may be used as part of an autoencoder, a spectral convolutional neural network (SCNN), a restricted Boltzmann machine model, another model described by Lee et al. (Lee, J., Kim, H., Lee, J. and Yoon, S., 2017, February. Transfer Learning for Deep Learning on Graph-Structured Data. In AAAI (pp. 2154-2160)), which is herein incorporated by reference.
In some embodiments, the process 4200 may include determining a second set of features for the machine learning model based on the log of states, as indicated by block 4220. The second set of features may include other attributes of the directed graph or other log of states described above, such as an identifier for a graph portion template matching the directed graph portion. Alternatively, or in addition, the second set of features may also include computed values determined from the attributes or data associated with a directed graph, where the computed values are not part of the first set of features. For example, some embodiments may determine a set of summations based on transaction scores associated with the directed graph and include the set of summations in a second set features of the directed graph in response to a determination that a first machine learning model should be used in block 4212.
Some embodiments may determine one or more features of the second set of features based on a selected learning model, such the machine learning model selected above for block 4212. Furthermore, as described further below, some embodiments may select different sets of parameter models based on different sets of features, even for sets of features that share a machine learning model. For example, in response to the selection of a first machine learning model, some embodiments may determine a first set of spectral values as part of a second set of features associated with a directed graph, where operations to determine a set of spectral values are described further below, where the set of spectral values may be used to select a set of model parameters for the machine learning model. In response to the selection of the same learning model but the determination of a different second set of features based on a different directed graph, some embodiments may then select a different set of model parameters.
In some embodiments, a feature of a directed graph encoding one or more values of a self-executing protocol may include a directed graph portion template generated from a directed graph using one or more operations described above. Alternatively, or in addition, determining a feature of a directed graph may include determining a new graph portion that removes one or more elements of the directed graph, where the new graph portion may be a subgraph of the directed graph. For example, some embodiments may generate a graph feature using one or more operations described by Bartha et al. (Bartha, M. and Krész, M., 2009, September. A depth-first algorithm to reduce graphs in linear time. In 2009 11th International Symposium on Symbolic and Numeric Algorithms for Scientific Computing (pp. 273-281). IEEE), which is hereby incorporated by reference. For example, some embodiments may recursively shrink a three-vertices chain of a directed graph portion to a two-vertices chain, where a three-vertices chain may include a first, second, and third vertex connected as a chain via directed edges, and use the directed graph portion having the two-vertices chain as a feature.
As described above, a feature of a directed graph may include values in a non-Euclidean domain as parameters for use in a graph convolutional neural network, where the graph convolutional neural network may be an SCNN. To determine a set of spectral values for an SCNN, some embodiments may use operations described by Ma et al. (Ma, Y., Hao, J., Yang, Y., Li, H., Jin, J. and Chen, G., 2019. Spectral-based graph convolutional network for directed graphs. arXiv preprint arXiv:1907.08990), which is hereby incorporated by reference. Some embodiments may determine a spectral value by determining or approximating an eigendecomposition of a Laplacian of a directed graph using Chebyshev polynomials. For example, some embodiments may determine an approximated transition probability matrix {tilde over (P)} indicating the probability that the state of a directed graph stored in program state may transition from a first state to a second state, determine a normalized Perron vector {tilde over (ϕ)}norm of {tilde over (P)}, or determine a diagonal of {tilde over (ϕ)}norm as {tilde over (Φ)}. Various approximation operations may be used. For example, some embodiments may determine {tilde over (P)} as based on Statement 15 below, equal to, where D may be the diagonal matrix representing node degrees of the directed graph and A may represent an adjacency matrix representing the directed graph:
{tilde over (P)}=D−1A (15)
In some embodiments, matrix operations such as matrix inversion or matrix multiplication may be implemented by a function or routine as a combination of element-wise summation, element-wise multiplication, or the like. Some embodiments determine an approximated Laplacian of the directed graph using {tilde over (P)}, {tilde over (Φ)}, or other values or sets of values. For example, some embodiments may determine a Laplacian L using Statement 16 below, where some embodiments may then determine a set of spectral values by performing a set of eigendecomposition operations on the Laplacian to determine the eigenvectors of the Laplacian:
L=I−½({tilde over (Φ)}1/2{tilde over (P)}{tilde over (Φ)}−1/2+{tilde over (Φ)}−1/2{tilde over (P)}T{tilde over (Φ)}1/2) (16)
Some embodiments receive a convolved form of the Laplacian or perform a convolution operation to determine the convolved form of the Laplacian, which may then be used to determine a directed graph convolution layer. Some embodiments may use one or more values of the directed graph convolution layer to determine a value. For example, some embodiments may perform operations similar to those used to implement Statement 17 below to determine Z, where X may be a feature vector for vertices of a directed graph, and where Ψ may be a filter parameter matrix:
Z=½({tilde over (Φ)}1/2{tilde over (P)}{tilde over (Φ)}−1/2+{tilde over (Φ)}−1/2{tilde over (P)}T{tilde over (Φ)}1/2)XΨ (17)
In some embodiments, the second set of features may be orthogonal to the first set of features, where orthogonal features may be independent of each other in a feature space. For example, the first set of features may include the presence of graph portion templates and the value of transaction scores, and the second set of features may include an amount of resources allocated to entities listed by a directed graph. By using of orthogonal sets of features to select a set of model parameter values, some embodiments may increase parameter selection accuracy when selecting a set of model parameter values.
Some embodiments may obtain a set of model parameter values of the machine learning model based on the first or second set of features, as indicated by block 4230. The set of model parameter values may include weights, biases, other stored values for neurons of a neural network, hyperparameters of a machine learning model, or the like. For example, some embodiments may obtain a first set of weights as model parameters for an SCNN based on the SCNN model being selected as a machine learning model and based on a second set of feature values being associated with the first set of weights. As described elsewhere in this disclosure, some embodiments may store model parameter values in association with features associated with a directed graph (e.g., via categories, parameters, or values of the directed graph), a graph portion template, or the like. In some embodiments, the set of model parameter values may be stored in association with a set of graph portion templates such as the library of graph portion templates described above.
As described above, some embodiments may use a first set of features such as specific vertices, vertex types, or graph portion templates to select a set of model parameters for a machine learning model. For example, some embodiments may perform a search through a database of model parameters using a set of features that includes a set of graph portion template identifiers as an index or search filter to retrieve a first set of model parameter values. Alternatively, or in addition, some embodiments may search for a set of model parameters based on a machine learning model determined based on the first set of features or a second set of features determined based on the machine learning model. Moreover, because specific vertices, vertex types, or graph portion templates are able to encode attributes of vertex types or graph portions, some embodiments may select specific sets of learning models or model parameters in response to the presence of a specific type of vertex or graph portion in a directed graph. For example, some embodiments may select a first set of model parameter values based on the first set of features indicating that a specific type of graph portion is present in a directed graph and may select second set of model parameter values based on the first set of features not indicating that the specific type of graph portion is present.
Some embodiments may obtain the set of model parameter values from a tamper-evident, distributed ledger executing on a peer-to-peer network of computing nodes. For example, some embodiments may obtain the set of model parameter values from a distributed database such as IFPS. Alternatively, or in addition, some embodiments may obtain the set of model parameter values from database stored on a cloud computing resource, such as a repository stored on an Amazon AWS server, Google Cloud server, or Microsoft Azure server.
Some embodiments may provide, from a first computer system, the set of model parameter values to a second computer system to perform a training operation, as indicated by block 4234. The second computer system may be a second computer system distinct from the first computer system described above. For example, the first computer system may be used to perform one or more operations of blocks 4204 to 4230, and the second computer system may be used to perform one or more operations described for block 4234. The address associated with the entity may include an API such as a web service address, an API specific to a program executing on a client computing device controlled by an entity (e.g., via a user credentialed as representing the entity), a text message for a mobile computing device, or the like. In some embodiments, an address associated with the entity may include an address of a peer computing node registered to the entity, where the node may then perform one or more operations using the set of model parameter values.
In some embodiments, receiving a set of model parameter values may cause a client computing device to train a version of the machine learning model corresponding to the set model parameter values or to otherwise use the set of model parameter values. For example, some embodiments may send a set of weights corresponding with five layers of an SCNN to a remote computing device as a web message. Receiving the web message may cause the remote computing device to use the set of weights to train a version of the SCNN to determine additional model parameter values for a seven-layer version of an SCNN or to directly determine an outcome score using the set of weights.
Some embodiments may send the set of model parameter values to a sub-application of a distributed application or otherwise perform additional training operations to determine an outcome score. For example, some embodiments may send the set of model parameter values to a specific node of a peer-to-peer network and perform additional operations to determine interpretable elements of directed graph data when determining an outcome score based on the directed graph. Various interpretability operations may be performed, such as an operation described by Carvalho et al. (Carvalho, D.V., Pereira, E. M. and Cardoso, J. S., 2019. Machine learning interpretability: A survey on methods and metrics. Electronics, 8(8), p. 832), which is hereby incorporated by reference.
While the above describes some embodiments as sending a stored set of model parameters, some embodiments may receive a first set of model parameters and perform additional training operations using the first set of model parameters. For example, some embodiments may receive a set of model parameters from a repository of model parameters via a message sent over a file transfer protocol (FTP). Furthermore, some embodiments may obtain a first set of model parameters from a publicly-accessible repository such as GitHub, a government database, or the like and perform a set of training operations using the first set of model parameters.
As described above, obtaining a trained machine learning model may cause or allow a computer system to perform additional training using the trained machine learning model. For example, after training the first machine learning model on a first training set with a first computer system controlled by a first entity, some embodiments may provide the first machine learning model to a second computer system controlled by a second entity. In some embodiments, the second entity does not permission to have access to at least some of the first training set used train the first machine learning model. The second computer system may obtain a second training set different from the first training set, where the second training set may include its own pairs of pairs of symbolic AI models and logs of state of the symbolic AI models. For example, the second training set may include a set of directed graphs representing logs of state for smart contracts used by the second entity, where the second training set is different from the first training set with respect at least one log of state of the second training set.
The second machine learning model may include the first machine learning model or parameters of the first machine learning model. As described further below for block 4328, training may include iteratively adjusting model parameters such as weights, biases, or other values of the machine learning model based on an objective function, where the objective function used to determine the second machine learning model may be the same as or different from an objective function used to determine the first machine learning model. When performing a second training operation to determine the second machine learning model, some embodiments may initialize parameters of the second machine learning model to have the same value as that of the parameters of the first machine learning model. After training the second machine learning model, some embodiments may store the second machine learning model in memory using the second computer system. Once stored, either or both the first machine learning model or second machine learning model may provide a step for machine learning and may be distinct from a symbolic AI model with respect to the types of data being encoded or used in the machine learning model.
In some embodiments, after receiving a first machine learning model from a first entity via a first computer system, a second computer system may train a second machine learning model on a second training set using the first machine learning model. For example, the first machine learning model may be a sub-model of the second machine learning model, where the sub-model may be used to provide a first output value. The second computer system may include an error-correcting machine learning model as a second sub-model of the second machine learning model, where the error-correcting machine learning model may then be used to modify the first output value to determine a second output value. Parameters of the error-correcting machine learning model may be iteratively adjusted during training to optimize an objective function based on the second output value. For example, some embodiments may iteratively update the weights of a set of neural network layers of an error-correcting machine learning model to minimize a cost function based on the second output value during a training operation using a second training set. In some embodiments, the second training operations may be performed without updating values of the first machine learning model. For example, after receiving a first machine learning model having fifteen parameters, some embodiments may iteratively update the weights of an error-correcting machine-learning model without changing any of the fifteen parameters.
As described above, some embodiments may perform a set of interpretability operations to link a specific log of states to an outcome score or other model-predicted or model-categorized value. When performing interpretability operation for a machine learning model on a log of states such as a directed graph, some embodiments may perform one or more operations to interpret an outcome score by indicating a corresponding feature of the directed graph (e.g., a parameter of a conditional statement associated with a vertex of the directed graph). Some embodiments may implement one or more operations described by Ying et al. (Ying, Z., Bourgeois, D., You, J., Zitnik, M. and Leskovec, J., 2019. Gnnexplainer: Generating explanations for graph neural networks. In Advances in neural information processing systems (pp. 9244-9255)), which is hereby incorporated by reference. In some embodiments, determining an interpretable element may include determining that a first graph portion of the directed graph is a significant contributor to an outcome score of a directed graph when using a machine learning model. In some embodiments, determining that a graph portion is a significant contributor to an outcome score of a directed graph may include determining that using the machine learning model provides on the first graph portion provides an outcome score that is quantitatively similar to the outcome score of the directed graph (e.g., within a range of 50%, of the outcome score of the directed graph). Some embodiments may then indicate the first graph portion via highlighting, coloring, animating, or otherwise visually indicating the first graph portion in a UI. Alternatively, or in addition, some embodiments may send an indicator of the first graph portion to the first entity.
Some embodiments may generate a plurality of simulated directed graphs having fewer edges or fewer vertices than the directed graph (i.e., a subgraph of the directed graph) during an iterative optimization operation. For example, some embodiments may use a machine learning model with the set of model parameter values to determine a set of simulated outcome scores based on a set of simulated subgraphs generated during a loss value minimization operation. Some embodiments may use a neural network to determine a set of maps (sometimes called a ‘saliency map,“mask,” saliency mask,’ or the like), such as a first map usable to select a subgraph and a second map used to determine features (e.g., conditional statement parameters, categories, vertex statuses, or the like). For example, some embodiments may use a neural network to determine a plurality of simulated outcome scores when performing a minimization operation with the neural network to minimize a loss value. The loss value may be based on a difference value and an edge size loss, where the difference value may be equal to or otherwise based on the simulated outcome score and the actual outcome score, and where the edge size loss may be based on a number of edges such that increasing the count of edges of a simulated subgraph increases the losses. Some embodiments may determine a plurality of simulated directed graphs by updating the set of maps. For example, some embodiments may update the first map or the second map described above to determine a simulated directed graph and indicate a graph portion matching the simulated directed graph.
Some embodiments may determine one or more subgraphs of a directed graph by updating the values of a directed graph or a graph portion of the directed graph. For example, some embodiments may update different conditional statement parameters or other data associated with vertices of a directed graph to generate a plurality of different sets of features by updating a feature map multiple times during a set of optimization operations. During the set of optimization operations, some embodiments may determine a simulated outcome score for each set of the simulated sets of features using the same set of model parameter values and learning model used to determine the outcome score for the actual directed graph. Some embodiments may select a simulated outcome score of the set of simulated outcome scores determined above based on a difference between the selected simulated outcome score and the first outcome score, where the selected outcome score may be associated with with a minimized loss value. Some embodiments may then select a first graph portion corresponding to the set of features associated with the selected simulated outcome score and send an indicator of the first graph portion to an entity or cause a device to visually indicate the first graph portion in a UI.
As described elsewhere in this disclosure, various outcome types may be used or selected by a user. For example, an outcome score type may indicates whether a program modeled by the directed graph will risk concurrently activating a pair of vertices where a first vertex obligating an entity participating in the program to perform an action that will trigger a prohibited action. In some embodiments, an outcome score type may used in conjunction with a warning. For example, a computing device may use the set of model parameters to predict an outcome score indicating whether a resource allocation will be exhausted. Some embodiments may determine whether the outcome score satisfies a warning threshold and, in response to the outcome score satisfying the warning threshold, send a warning message to an entity responsible for allocating the resource, an observer entity, an entity indicated as receiving the resource, or another entity.
Some embodiments may obtain an updated set of model parameter values from the entity, as indicated by block 4240. Some embodiments may an obtain an entity-provided set of parameter values that are different from the set of model parameter values sent to the entity via the entity address. For example, some embodiments may have sent a set of weights corresponding to three layers of an SCNN learning model and obtained an entity-provided set of weights for four layers of an SCNN learning model, where the values of the entity-provided layers may correspond to one or more layers.
Some embodiments may implement one or more operations to practice differential privacy. Various differential privacy operations may be implemented, such as anonymizing identifying data or adding noise or otherwise fuzzifying numeric values. For example, some embodiments may, after obtaining an initial set of model parameter values, outcome scores, or other values, add or subtract a random value determined by a random or pseudorandom number generator update to the initial set of model parameter values, outcome scores, or other values. Some embodiments may aggregate multiple values of a similar type, such as the same model parameter type or same outcome score together into a single combined value, and store the single combined value for further analysis. For example, some embodiments may obtain a first set of model parameter values from a first client computing device and a second set of model parameter values from a second client computing device, where the first and second sets of model parameter values both correspond to a same learning model such as an SCNN. Some embodiments may then determine a mean average value (e.g., via a summation and division operation) or another measurement of centralized tendency as an aggregated value. As discussed elsewhere in this disclosure, some embodiments may use the aggregated value to determine or update a set of model parameter values.
Some embodiments may implement a parameter-updating operation based on model parameter values provided by entities. Some embodiments may, after obtaining the entity-provided set of model parameter values, use the entity-provided set of model parameter values in a machine learning model to determine an updated set of outcome scores. Some embodiments may compare the updated outcome score to an outcome score threshold based on a pre-existing outcome score or other threshold value. If the outcome score satisfies the outcome score threshold, some embodiments may update an improvement score or other value indicating that the entity-provided model parameter values contributed by the entity satisfy the outcome score threshold. For example, some embodiments may obtain an entity-provided set of learning model weights from a client computing device. Some embodiments may then use, via a remote server or a distributed computing platform, the entity-provided model parameter values for a testing operation with a version of the machine learning model. In response to a determination that a machine learning model using the entity-provided model parameter values provides an accuracy that is greater than a previous accuracy with respect to the testing operation result, some embodiments may replace or otherwise update the previous set of model parameter values with the entity-provided set of model parameter values.
In some embodiments, the process 4300 may include obtaining attribute(s) or other data associated with the set of logs of states, as indicated by block 4308. Attributes or other data of a set of logs of states may include categories associated with vertices or edges of the directed graph. For example, attributes or other data of a directed graph or other data representing a log of states may include categories such as “rights,” “obligations,” or “prohibitions,” as selected from the set of mutually-exclusive categories represented by the array ‘[“rights”; “obligations”; “prohibitions”]. In some embodiments, attributes or other data of a directed graph may include vertex statuses such as “satisfied,” “fulfilled,” “triggered,” “active,” “cancelled,” “inactive,” or the like. Various specific vocabulary may be used to indicate a vertex status, where the vertex status may represent a state indicating whether a conditional statement of a vertex is inactive (e.g., satisfying the conditional statement will not set the vertex to a different state), active (e.g., satisfying the conditional statement will set the vertex to a different state), triggered (e.g., the conditional statement of the vertex has been satisfied or a time threshold of the vertex has been reached without conditional statement being satisfied), or the like.
In some embodiments, attributes or other data of the directed graph may include an entity or entity role associated with a directed graph via a list or conditional statement. For example, some embodiments may obtain an entity role as an attribute associated with the directed graph based on the program state storing the directed graph also listing an entity having the entity role. Some embodiments may obtain attributes or data associated with the directed graph from a same memory storage used to store the directed graph itself. Alternatively, or in addition, some embodiments may obtain attributes or data associated with the directed graph from a source that is different from where the directed graph itself is stored. For example, some embodiments may obtain the directed graph from a tamper-evident, distributed ledger and obtain an entity role from a cloud computing resource, such as a cloud server.
In some embodiments, the process 4300 may include determining one or more sets of features based on the set of logs of state, as indicated by block 4320. A set of features may include attributes such as an identifier for a graph portion matching a graph portion template. Alternatively, or in addition, the attributes may include the presence or count of vertices having a specific vertex type. For example, the attributes may include a boolean indicating whether a “prohibition” vertex is active or a count of vertices labeled as “obligations” and having conditional statement parameters greater than a threshold value. Some embodiments may perform a BFS operation to search for each graph portion of each respective directed graph or other log of states that matches a first graph portion template or other stored pattern and store an identifier for the matching graph portion or other stored pattern in association with the directed graph. Alternatively, or in addition, the multiple sets of features may include computed features such as a spectral domain value, another value in a non-Euclidean representation of a feature domain, a measure of centralized tendency for a set of conditional statement parameters, or the like. Some embodiments may further store edges associating different graph portions, where each of the different graph portions matches with a different graph portion template. Additionally, in some embodiments, the smart contract graph features may include spectral values, where operations to determine the spectral values may use operations described for block 4220.
Some embodiments may determine multiple sets of features by determining a set of features for each directed graph of the plurality of directed graphs. For example, some embodiments may, for each respective directed graph, determine a respective set of features that include a set of spectral values, count of satisfied conditions associated with each category of a set of mutually-exclusive categories, a count of graph portions matching a graph portion template, or the like.
In some embodiments, a set of features may include a set of transaction scores. For example, some embodiments may obtain records of the transactions between entities recorded for each self-executing protocol and obtain a set of transaction scores of the transactions. For example, some embodiments may obtain transaction scores for transactions of a set of self-executing protocols, where the transaction scores may include amounts of resources being allocated, amounts of digital assets being exchanged, amounts of a currency being transferred, or the like. A transaction of a directed graph of a self-executing protocol may include a transaction that triggered a vertex of a directed graph, such as by satisfying a conditional statement of a vertex while a status of the vertex indicated that the vertex was active (e.g., satisfying a conditional statement of a prohibition vertex and causing the prohibition vertex to be indicated as ‘failed’). Alternatively, or in addition, a transaction of a directed graph of a self-executing protocol may include a transaction that activated a vertex of a directed graph (e.g., by satisfying a conditional statement that caused the vertex to become indicated as active). As discussed elsewhere in this disclosure, some embodiments may perform one or more differential privacy operations to increase the privacy of the entities associated with the transaction scores, such as aggregating (summing, determining a mean average, etc.) two or more scores together before storing or using the aggregated score to determine a set of model parameter values.
In some embodiments, the process 4300 may include determining a set of labels based on the one or more sets of features, as indicated by block 4324. Some embodiments may determine a label for a log of states based on an attribute, feature, or other data associated with the log of states. For example, some embodiments may determine that a program having program state storing the directed graph lists an entity having an entity role. Some embodiments may then associate the directed graph with the entity role and use the entity role as a label. Similarly, a label may be based on a category of a vertex, a conditional statement associated with a vertex, a conditional statement parameter (e.g., a threshold value, a threshold range, a category, or the like), a vertex status, whether a vertex having a specific category has been satisfied or failed, or the like. Some embodiments may determine that a graph portion of a directed matches with a graph portion template and, in response, associate the directed graph with a graph label based on the graph portion template. As further described below, associating a respective directed graph of the plurality of directed graphs with a set of labels may allow different subsets of directed graphs to be trained, where each subset of directed graphs may share at least one graph label of the set of labels or may otherwise be organized based on their respective set of labels.
In some embodiments, the process 4300 may include determining a set of model parameter values of a machine learning model using a training operation based on the one or more sets of features, as indicated by block 4328. Some embodiments may determine a set of model parameter values by training a machine learning model, such as a neural network, support vector machines, random forest model, or the like. For example, some embodiments may determine a set of model parameter values for a RNN. In some embodiments, the training of the machine learning model may include using features determined from the set of sequences of states described above as inputs for the machine learning model. For example, some embodiments may determine a set of model parameter values for an SCNN or other neural network based on a set of features including a set of spectral values of a directed graph.
Various operations may be performed during training to determine a set of model parameters for a machine learning model. For example, some embodiments may implement a backpropagation method to train a neural network using a log of states as training inputs, where some embodiments may update the weights of the neurons of a neural network using a gradient of an objective function of the training operation. In some embodiments, training may include optimizing a set of model parameters based on an objective function (e.g., maximizing a gain function, minimizing a loss function, or the like). For example, some embodiments may perform a first-order iterative optimization algorithm, such as gradient descent, to determine the model parameters of an RNN. Alternatively, or in addition, some embodiments may use other optimization methods to determine model parameters for a machine learning model, such as genetic algorithm methods, simulated annealing methods, or particle swarm optimization methods. For example, some embodiments may perform training operations for an RNN using fifty sequences of program states, each sequence corresponding to an evolution of state for fifty different smart contracts. Such training operations may be used for various learning models, such as an RNN, dynamic Bayesian network, hidden Markov model, random forest, or the like.
Various smoothing or activation functions may be used during training. For example, some embodiments may determine a set of neural network weights corresponding to a set of neurons of an SCNN network described by Ma et al. to train the SCNN by using a set of softmax functions and rectified linear units (ReLU) functions to determine neural network weights W(0) and W(1), such as by implementing a set of operations to solve for the convolution layer Z in Statement 18 below, where  is defined by Statement 19, {tilde over (P)} is a transition probability matrix, {tilde over (Φ)} is a diagonal of a normalized Perron vector of a directed graph, and X is a vector of vertex features:
Z=softmax( ReLU(Â×W(0))W(1)) (18)
A=½({tilde over (Φ)}1/2{tilde over (P)}{tilde over (Φ)}−1/2+{tilde over (Φ)}−1/2{tilde over (P)}T{tilde over (Φ)}1/2) (19)
Some embodiments may perform the training using a set of training outputs, where the set of training outputs may be obtained from a database of historical outcome scores. In some embodiments, the training outputs may include values stored in the database of historical outcome scores or computed from the values stored in the database. Some embodiments may obtain different sets of training outputs based on different target training categories. For example, some embodiments may perform a training operation to determine a first set of model parameter values using a first set of training outputs, where the first set of training outputs may be selected based on a target training category being set to a first category. Some embodiments may perform multiple training operations to determine different sets of model parameters using different sets of training outputs, where some or all of the multiple training operations may share a same set of directed graphs or features determined from the set of directed graphs.
As discussed elsewhere in this disclosure, some embodiments may train different versions of related machine learning models using different computer systems. After obtaining a plurality of pairs of symbolic AI models and logs of state of the symbolic AI models for use a first training set with a first computer system controlled by a first entity, some embodiments may train a first machine learning model with the training set using one or more operations in this disclosure. For example, some embodiments may train a RNN by iteratively adjusting weights, biases, or other model parameters of the RNN based on an objective function during a first training operation. Once the first training operation is completed, the set of model parameters can be provided to a second computer system for directly determining outcome scores or other predictions or for or use in additional training operations, such as operations described for block 4234.
In some embodiments, the process 4300 may include storing a set of model parameter values in association with the machine learning model or labels, as indicated by block 4332. One or more data structures may be used to store the set of model parameter values. For example, some embodiments may store some or all of the set of model parameter values in the library of graph portion templates, where a record for a graph portion template may include a set or subset of model parameter values associated with a graph portion template. Alternatively, or in addition, some embodiments may store some or all of the set of model parameter values in a database indexed by one or more of the labels or identifiers of a machine learning model. For example, some embodiments may store the set of model parameter values in a SQL database associated with an SCNN an indexed by an index value that encodes one or more labels such that a query indicating an identifier for a spectral convolutional neural network as a machine learning model and a specific set of labels will provide the specific set of model parameter values. Additionally, some embodiments may select a set of model parameter values based on a set of conditional statements or a set of conditional statement parameters. For example, some embodiments may select a first set of model parameters based on a first conditional statement parameter indicating that the resource type of a corresponding first conditional statement is “monthly allocated memory.” Some embodiments may also select a second set of model parameters based on a second conditional statement parameter indicating that the resource type of a corresponding second conditional statement is “monthly allocated processor time.”
The set of model parameter values may be stored on a cloud computing resource, a single node or subset of nodes of a peer-to-peer network, a tamper-evident, distributed ledger distributed across the peer-to-peer network, or the like. As described above, some embodiments may update one or more values of a set of model parameter values based on an updated set of model parameter values provided by the entity. For example, some embodiments may replace an initial set of model parameter values with a new set of model parameter values provided by an entity. Alternatively, or in addition, some embodiments may add a new set of model parameter values for a new layer of the neural network. Alternatively, or in addition, some embodiments may obtain a new learning model via a message including or associated with the updated set of model parameter values.
In block diagrams, illustrated components are depicted as discrete functional blocks, but embodiments are not limited to systems in which the functionality described herein is organized as illustrated. The functionality provided by each of the components may be provided by software or hardware modules that are differently organized than is presently depicted, for example such software or hardware may be intermingled, conjoined, replicated, broken up, distributed (e.g. within a data center or geographically), or otherwise differently organized. The functionality described herein may be provided by one or more processors of one or more computers executing code stored on a tangible, non-transitory, machine readable medium. In some cases, notwithstanding use of the singular term “medium,” the instructions may be distributed on different storage devices associated with different computing devices, for instance, with each computing device having a different subset of the instructions, an implementation consistent with usage of the singular term “medium” herein. In some cases, third party content delivery networks may host some or all of the information conveyed over networks, in which case, to the extent information (e.g., content) is said to be supplied or otherwise provided, the information may provided by sending instructions to retrieve that information from a content delivery network.
The reader should appreciate that the present application describes several independently useful techniques. Rather than separating those techniques into multiple isolated patent applications, applicants have grouped these techniques into a single document because their related subject matter lends itself to economies in the application process. But the distinct advantages and aspects of such techniques should not be conflated. In some cases, embodiments address all of the deficiencies noted herein, but it should be understood that the techniques are independently useful, and some embodiments address only a subset of such problems or offer other, unmentioned benefits that will be apparent to those of skill in the art reviewing the present disclosure. Due to costs constraints, some techniques disclosed herein may not be presently claimed and may be claimed in later filings, such as continuation applications or by amending the present claims. Similarly, due to space constraints, neither the Abstract nor the Summary of the Invention sections of the present document should be taken as containing a comprehensive listing of all such techniques or all aspects of such techniques.
It should be understood that the description and the drawings are not intended to limit the present techniques to the particular form disclosed, but to the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present techniques as defined by the appended claims. Further modifications and alternative embodiments of various aspects of the techniques will be apparent to those skilled in the art in view of this description. Accordingly, this description and the drawings are to be construed as illustrative only and are for the purpose of teaching those skilled in the art the general manner of carrying out the present techniques. It is to be understood that the forms of the present techniques shown and described herein are to be taken as examples of embodiments. Elements and materials may be substituted for those illustrated and described herein, parts and processes may be reversed or omitted, and certain features of the present techniques may be utilized independently, all as would be apparent to one skilled in the art after having the benefit of this description of the present techniques. Changes may be made in the elements described herein without departing from the spirit and scope of the present techniques as described in the following claims. Headings used herein are for organizational purposes only and are not meant to be used to limit the scope of the description.
As used throughout this application, the word “may” is used in a permissive sense (i.e., meaning having the potential to), rather than the mandatory sense (i.e., meaning must). The words “include”, “including”, and “includes” and the like mean including, but not limited to. As used throughout this application, the singular forms “a,” “an,” and “the” include plural referents unless the content explicitly indicates otherwise. Thus, for example, reference to “an element” or “a element” includes a combination of two or more elements, notwithstanding use of other terms and phrases for one or more elements, such as “one or more.” The term “or” is, unless indicated otherwise, non-exclusive, i.e., encompassing both “and” and “or.” Terms describing conditional relationships, e.g., “in response to X, Y,” “upon X, Y,”, “if X, Y,” “when X, Y,” and the like, encompass causal relationships in which the antecedent is a necessary causal condition, the antecedent is a sufficient causal condition, or the antecedent is a contributory causal condition of the consequent, e.g., “state X occurs upon condition Y obtaining” is generic to “X occurs solely upon Y” and “X occurs upon Y and Z.” Such conditional relationships are not limited to consequences that instantly follow the antecedent obtaining, as some consequences may be delayed, and in conditional statements, antecedents are connected to their consequents, e.g., the antecedent is relevant to the likelihood of the consequent occurring. Statements in which a plurality of attributes or functions are mapped to a plurality of objects (e.g., one or more processors performing steps A, B, C, and D) encompasses both all such attributes or functions being mapped to all such objects and subsets of the attributes or functions being mapped to subsets of the attributes or functions (e.g., both all processors each performing steps A-D, and a case in which processor 1 performs step A, processor 2 performs step B and part of step C, and processor 3 performs part of step C and step D), unless otherwise indicated. Similarly, reference to “a computer system” performing step A and “the computer system” performing step B can include the same computing device within the computer system performing both steps or different computing devices within the computer system performing steps A and B. Further, unless otherwise indicated, statements that one value or action is “based on” another condition or value encompass both instances in which the condition or value is the sole factor and instances in which the condition or value is one factor among a plurality of factors. Unless otherwise indicated, statements that “each” instance of some collection have some property should not be read to exclude cases where some otherwise identical or similar members of a larger collection do not have the property, i.e., each does not necessarily mean each and every. Limitations as to sequence of recited steps should not be read into the claims unless explicitly specified, e.g., with explicit language like “after performing X, performing Y,” in contrast to statements that might be improperly argued to imply sequence limitations, like “performing X on items, performing Y on the X'ed items,” used for purposes of making claims more readable rather than specifying sequence. Statements referring to “at least Z of A, B, and C,” and the like (e.g., “at least Z of A, B, or C”), refer to at least Z of the listed categories (A, B, and C) and do not require at least Z units in each category. Unless specifically stated otherwise, as apparent from the discussion, it is appreciated that throughout this specification discussions utilizing terms such as “processing,” “computing,” “calculating,” “determining” or the like refer to actions or processes of a specific apparatus, such as a special purpose computer or a similar special purpose electronic processing/computing device. Features described with reference to geometric constructs, like “parallel,” “perpendicular/orthogonal,” “square”, “cylindrical,” and the like, should be construed as encompassing items that substantially embody the properties of the geometric construct, e.g., reference to “parallel” surfaces encompasses substantially parallel surfaces. The permitted range of deviation from Platonic ideals of these geometric constructs is to be determined with reference to ranges in the specification, and where such ranges are not stated, with reference to industry norms in the field of use, and where such ranges are not defined, with reference to industry norms in the field of manufacturing of the designated feature, and where such ranges are not defined, features substantially embodying a geometric construct should be construed to include those features within 15% of the defining attributes of that geometric construct. The term “set” may indicate a single item or a plurality of items, e.g., “set of widgets” may indicate only one widget or may indicate multiple widgets. The terms “first”, “second”, “third,” “given” and so on, if used in the claims, are used to distinguish or otherwise identify, and not to show a sequential or numerical limitation. As is the case in ordinary usage in the field, data structures and formats described with reference to uses salient to a human need not be presented in a human-intelligible format to constitute the described data structure or format, e.g., text need not be rendered or even encoded in Unicode or ASCII to constitute text; images, maps, and data-visualizations need not be displayed or decoded to constitute images, maps, and data-visualizations, respectively; speech, music, and other audio need not be emitted through a speaker or decoded to constitute speech, music, or other audio, respectively. Computer implemented instructions, commands, and the like are not limited to executable code and can be implemented in the form of data that causes functionality to be invoked, e.g., in the form of arguments of a function or API call.
In this patent, to the extent any U.S. patents, U.S. patent applications, or other materials (e.g., articles) have been incorporated by reference, the text of such materials is only incorporated by reference to the extent that no conflict exists between such material and the statements and drawings set forth herein. In the event of such conflict, the text of the present document governs, and terms in this document should not be given a narrower reading in virtue of the way in which those terms are used in other materials incorporated by reference.
The present techniques will be better understood with reference to the following enumerated embodiments:
A-1. A tangible, non-transitory, machine-readable medium storing instructions that when executed by one or more processors effectuate operations comprising: determining, with a computer system, that an event has occurred; selecting, with the computer system, a self-executing protocol among a plurality of self-executing protocols based on the event, wherein: the self-executing protocol comprises a set of conditions, a set of entities, a set of vertices, and a set of directed graph edges connecting the set of vertices, the set of vertices comprise different respective subsets of the conditions, the set of entities are encoded in an associative array, the set of conditions are encoded in an associative array, the set of vertices are encoded as a serialized array of vertices, wherein the serialized array of vertices is in a serialized data format in persistent storage, selecting is based on whether the event satisfies any of the set of conditions; deserializing, with the computer system, the serialized array of vertices to generate a directed graph in a non-persistent memory, wherein the directed graph encodes the set of conditions, set of vertices, set of entities, and set of directed edges; determining, with the computer system, a set of triggerable vertices from the vertices of the directed graph in the non-persistent memory; determining, with the computer system, a set of triggered vertices from the set of triggerable vertices based on which of the set of triggerable vertices are associated with the set of conditions satisfied by the event; updating, with the computer system, the directed graph in the non-persistent memory based on the set of triggered vertices, wherein updating the directed graph comprises, for each respective triggered vertex of the set of triggered vertices: updating a first value associated with the respective triggered vertex based on the event, where the first value indicates whether the respective triggered vertex is triggerable; updating a respective adjacent vertex to indicate that the respective adjacent vertex is triggerable, wherein the respective adjacent vertex is associated with a directed graph edge of the respective triggered vertex; updating, with the computer system, the serialized array of vertices by serializing the directed graph in the non-persistent memory after updating the directed graph in the non-persistent memory based on the set of triggered vertices; and persisting, with the computer system, the serialized array of vertices to the persistent storage after the serialized array of vertices is updated by serialization.
A-2. The medium of embodiment A-1, wherein: a first vertex in the set of vertices is indicated to not be triggerable by a first set of values, wherein each of the first set of values indicate whether a vertex in the set of vertices is triggerable; and the directed graph in the non-persistent memory does not include the first vertex of the serialized array of vertices.
A-3. The medium of any of embodiments A-1 to A-2, wherein the serialized array of vertices comprises an array of subarrays, wherein each subarray comprises a head vertex of a directed graph edge of the set of directed graph edges, a tail vertex of the directed graph edge, a label associated with the directed graph edge, and a valence value indicating a number of other edges associated with the directed graph edge.
A-4. The medium of any of embodiments A-1 to A-3, wherein determining that an event occurred further comprises: receiving an event message from a publisher, wherein the publisher is identified by a publisher identifier; determining whether the publisher is associated with one of a set of authorized publishers based on the publisher identifier; and authorizing the event message based on a determination that the publisher identifier is associated with one of the set of authorized publishers.
A-5. The medium of any of embodiments A-1 to A-4, wherein the operations further comprise: receiving an event message from a publisher, wherein the event message is associated with a signature value and a publisher identifier; retrieving a cryptographic certificate based on the publisher identifier; computing a cryptographic hash value based on the signature value; and authenticating the event message based on the cryptographic hash value and the cryptographic certificate.
A-6. The medium of any of embodiments A-1 to A-5, wherein determining the set of triggered vertices comprises: determining a first set of vertices in the directed graph in the non-persistent memory, wherein each respective vertex of the first set of vertices is indicated as a head vertex by one of the set of directed graph edges; and determining the set of triggerable vertices based on the first set of vertices by filtering out a set of tail vertices from the first set of vertices, wherein each of the set of tail vertices is indicated as a tail vertex by one of the set of directed graph edges.
A-7. The medium of any of embodiments A-1 to A-6, wherein the serialized array of vertices is stored in a tamper-evident data store being executed by a set of peer nodes, wherein the tamper-evident data store comprises a directed acyclic graph of cryptographic hash pointers, and wherein deserializing the serialized array of vertices comprises using a first node of the set of peer nodes to deserialize the serialized array of vertices, and wherein the operations further comprising transmitting the serialized array of vertices from the first node to another node of the set of peer nodes after updating the serialized array of vertices.
A-8. The medium of any of embodiments A-1 to A-7, the operations further comprising receiving an event message, wherein receiving the event message comprises receiving a request that comprises the event message, and wherein the request comprises a method identifier and a host identifier, wherein the method identifier indicates that the request comprises an amount of data to modify data stored by the system, and wherein the host identifier indicates a host of the self-executing protocol.
A-9. The medium of any of embodiments A-1 to A-8, the operations further comprising receiving an event message, wherein the event message comprises a routing key, and wherein a data broker stores the event message in a queue, and wherein a protocol broker transmits the event message to an API associated with the self-executing protocol based on the routing key.
A-10. The medium of any of embodiments A-1 to A-9, wherein determining the set of triggered vertices comprises determining the set of triggered vertices based on a second set of values, wherein each of the second set of values is associated with one of a set of vertices of the directed graph in the non-persistent memory, and wherein one of the second set of values indicate that one of the set of vertices of the directed graph in the non-persistent memory is triggerable.
A-11. The medium of any of embodiments A-1 to A-10, wherein determining that the event has occurred comprises determining that a condition expiration threshold has been satisfied, and wherein the condition expiration threshold is associated with a first condition of a first triggerable vertex, and wherein the event does not satisfy the first condition.
A-12. The medium of any of embodiments A-1 to A-11, the operations further comprising updating an array of previously-triggered vertices based on a vertex identifier associated with the respective triggered vertex.
A-13. The medium of any of embodiments A-1 to A-12, the operations further comprising generating an initial directed graph based on an initial set of vertices, wherein the initial set of vertices is different from the serialized array of vertices.
A-14. The medium of any of embodiments A-1 to A-13, wherein a vertex of the directed graph stored in the non-persistent memory comprises a condition of the set of conditions.
A-15. The medium of any of embodiments A-1 to A-14, the operations further comprising updating a third set of values associated with the serialized array of vertices, wherein the third set of values indicate that the respective triggered vertex is not triggerable.
A-16. The medium of any of embodiments A-1 to A-15, wherein updating the respective adjacent vertex comprises setting a plurality of statuses associated with a plurality of vertices other than the respective triggered vertex as not triggerable.
A-17. The medium of any of embodiments A-1 to A-16, wherein updating the first value comprises updating the first value to indicate that the respective triggered vertex remains triggerable after updating the serialized array of vertices.
A-18. The medium of embodiment A-17, wherein updating the respective adjacent vertex comprises decreasing a second value, wherein the second value indicates a state of the self-executing protocol.
A-19. The medium of any of embodiments A-1 to A-18, the operations further comprising updating a set of previous events based on the event, wherein the set of previous events comprises a plurality of previous events that caused a state change in the self-executing protocol, wherein the set of previous events comprises a time during which the event occurred.
A-20. A tangible, non-transitory, machine-readable medium storing instructions that when executed by one or more processors effectuate operations comprising: determining, with a computer system, that an event has occurred; selecting, with the computer system, a self-executing protocol among a plurality of self-executing protocols based on the event, wherein: the self-executing protocol comprises a set of conditions, a set of entities, a set of vertices, and a set of directed graph edges connecting the set of vertices, the set of vertices comprise different respective subsets of the conditions, the set of entities are encoded in an associative array, the set of conditions are encoded in an associative array, the set of vertices are encoded as a serialized array of vertices, wherein the serialized array of vertices is in a serialized data format in persistent storage, selecting is based on whether the event satisfies any of the set of conditions; deserializing, with the computer system, the serialized array of vertices to generate a directed graph in a non-persistent memory, wherein the directed graph encodes the set of conditions, set of vertices, set of entities, and set of directed edges; determining, with the computer system, a set of triggerable vertices from the vertices of the directed graph in the non-persistent memory; determining, with the computer system, a set of triggered vertices from the set of triggerable vertices based on which of the set of triggerable vertices are associated with the set of conditions satisfied by the event; updating, with the computer system, the directed graph in the non-persistent memory based on the set of triggered vertices, wherein updating the directed graph comprises, for each respective triggered vertex of the set of triggered vertices: updating a first value associated with the respective triggered vertex based on the event, where the first value indicates whether the respective triggered vertex is triggerable; updating a respective adjacent vertex to indicate that the respective adjacent vertex is triggerable, wherein the respective adjacent vertex is associated with a directed graph edge of the respective triggered vertex; updating, with the computer system, the serialized array of vertices by serializing the directed graph in the non-persistent memory after updating the directed graph in the non-persistent memory based on the set of triggered vertices; and persisting, with the computer system, the serialized array of vertices to the persistent storage after the serialized array of vertices is updated by serialization.
A-21. A method to perform the operations of any of the embodiments A-1 to A-19.
A-22. A system, comprising: one or more processors; and memory storing instructions that when executed by the processors cause the processors to effectuate operations comprising: the operations of any one of embodiments A-1 to A-19.
B-1. A tangible, non-transitory, machine-readable medium storing instructions that when executed by one or more processors effectuate operations comprising: obtaining, with a computer system, a set of conditional statements, wherein: a conditional statement of the set of conditional statements is associated with an outcome subroutine that specifies operations in each of one or more branches of the conditional statement, a set of index values index the set of conditional statements, and a first outcome subroutine of a first conditional statement of the set of conditional statements uses a first index value of the set of index values, wherein the first index value is associated with a second conditional statement of the set of conditional statements; executing, with the computer system, a program instance of an application based on the set of conditional statements, wherein program state data of the program instance comprises: a set of vertices and a set of directed graph edges, wherein each of the set of vertices comprises a identifier value and is associated with one of the set of conditional statements, and wherein each of the set of directed graph edges associates a pair of the set of vertices and a direction from a tail vertex of the pair to a head vertex of the pair, a set of statuses, wherein each of the set of statuses is associated with one of the set of vertices, a set of vertex categories, wherein each of the set of vertex categories is a category value and is associated with a respective vertex of the set of vertices and is determined based a respective conditional statement of the respective vertex, and a set of scores, wherein each respective score of the set of scores is associated with a respective vertex and is based a respective conditional statement of the respective vertex; updating, with the computer system, the program state data based on a set of inputs comprising a first input, wherein updating the program state data comprises: modifying a status of a first vertex of the set of vertices based on the first input, updating a vertex adjacent to the first vertex; and determining, with the computer system, an outcome score based on the set of scores after updating the program state data.
B-2. The medium of embodiment B-1, wherein the status is a first status, and wherein updating the program state data comprises updating the program state data based on the first status, and wherein the operations further comprise: modifying a second status of a second vertex of the set of vertices based on a second input; updating a third vertex adjacent to the second vertex, wherein determining the outcome score comprises determining the outcome score after updating the third vertex.
B-3. The medium of embodiment B-2, wherein the operations further comprise determining the first input based on a probability value associated with one of the set of vertex categories.
B-4. The medium of any of embodiments B-2 to B-3, wherein the outcome score is a first outcome score, and wherein the program state data is in a first state before modifying the program state data, and wherein the operations further comprise: updating a neural network parameter after updating the third vertex based on the first outcome score, wherein the neural network parameter comprises a set of probability values assigned to each of a subset of vertices of the set of vertices; determining a third input based on the neural network parameter; updating the program state data that is in the first state based on the third input; and determining a second outcome score after updating the program state data based on the third input.
B-5. The medium of any of embodiments B-1 to B-4, wherein executing the program instance comprises executing the program instance during a first iteration, and wherein the set of inputs is a first set of inputs, and wherein the outcome score is a first outcome score, and wherein the program state data is in a first state before modifying the program state data, and wherein the operations further comprise: executing the program instance during a second iteration by updating the program state data based on a second set of inputs, wherein the program state data is in the first state before updating the program state data based on the second set of inputs; determining a second outcome score based on the second set of inputs; and determining a multi-iteration score based on the first outcome score and the second outcome score.
B-6. The medium of embodiment B-5, wherein the operations further comprise: acquiring a third score; and determining a possible event based the third score using a probability distribution, wherein the probability distribution is based on the multi-iteration score.
B-7. The medium of embodiment B-6, wherein determining the possible event comprises using a neural network that is trained using inputs based on the first outcome score and the second outcome score, and wherein the neural network is trained using a training output based on the first set of inputs and the second set of inputs.
B-8. The medium of any of embodiments B-5 to B-7, wherein: the first set of inputs is associated with a first weighting value; the second set of inputs is associated with a second weighting value; and determining the multi-iteration score is based on the first weighting value and the second weighting value.
B-9. The medium of any of embodiments B-5 to B-8, the operations further comprising determining a probability distribution function based on the multi-iteration score.
B-10. The medium of any of embodiments B-1 to B-9, wherein modifying the status of the first vertex comprises determining a set of events, wherein each of the set of events satisfies a condition of the set of conditional statements.
B-11. The medium of any of embodiments B-1 to B-10, wherein acquiring the set of conditional statements comprises: acquiring an event; for a respective self-executing protocol of a plurality of self-executing protocols, determining whether the event satisfies a condition associated with the respective self-executing protocol; and acquiring the set of conditional statements associated with the respective self-executing protocol in response to the event satisfying the condition associated with the respective self-executing protocol.
B-12. The medium of any of embodiments B-1 to B-11, wherein acquiring the set of conditional statements comprises: acquiring an entity identifier; for a respective self-executing protocol of a plurality of self-executing protocols, determining whether the entity identifier is in a respective set of entities associated with the respective self-executing protocol; and acquiring the set of conditional statements associated with the respective self-executing protocol in response to the entity identifier being in the respective set of entities associated with the respective self-executing protocol.
B-13. The medium of any of embodiments B-1 to B-12, the operations further comprising: acquiring a first entity identifier and a second entity identifier; selecting a first set of self-executing protocols from a plurality of self-executing protocols, wherein each of the first set of self-executing protocols comprises a first set of entities that comprises the first entity identifier; determining a second set of self-executing protocols from the plurality of self-executing protocols, wherein each of the second set of self-executing protocols comprises a second set of entities that comprises the second entity identifier; and determining a set of intermediary entities, wherein each of the set of intermediary entities is in a set of entities of the first set of self-executing protocols, and wherein each of the set of intermediary entities is in a set of entities of the second set of self-executing protocols.
B-14. The medium of any of embodiments B-1 to B-13, wherein modifying the status of the first vertex comprises setting a first status to indicate that a first entity fails to transfer a score to a second entity.
B-15. The medium of any of embodiments B-1 to B-14, the operations further comprising: detecting a pattern based on a plurality of the set of vertices and a plurality of the set of directed graph edges; and sending a message indicating that the pattern is detected.
B-16. The medium of any of embodiments B-1 to B-15, the operations further comprising determining a measure of central tendency based on the outcome score.
B-17. The medium of any of embodiments B-1 to B-16, the operations further comprising determining a kurtosis value based on the outcome score, wherein the kurtosis value correlates with a ratio of a first value and a second value, wherein the first value is based on a measure of central tendency, and wherein the second value is based on a measure of dispersion.
B-18. The medium of any of embodiments B-1 to B-17, the operations further comprising: acquiring an event message via an application protocol interface; determining a first set of events based on the event message, wherein the set of inputs does not include the first set of events; and updating the program state data based on the first set of events, wherein the program state data is updated based on the set of inputs after the program state data is updated with the first set of events.
B-19. The medium of any of embodiments B-1 to B-18, the operations further comprising: modifying a first status of a first vertex of the set of vertices to indicate that the first vertex is triggered; modifying a second status of a second vertex of the set of vertices to indicate that the second vertex is triggered; and in response to the first status and the second status being modified to indicate they are triggered, triggering a third vertex that is adjacent to the first vertex and the second vertex.
B-20. A method comprising: acquiring a set of conditional statements, wherein: a conditional statement of the set of conditional statements is associated with an outcome subroutine and an index value of a set of index values, and a first outcome subroutine of a first conditional statement of the set of conditional statements uses a first index value of the set of index values, wherein the first index value is associated with a second conditional statement of the set of conditional statements; executing a program instance of an application based on the set of conditional statements, wherein program state data of the program instance comprises: a set of vertices and a set of directed graph edges, wherein each of the set of vertices comprises a identifier value and is associated with one of the set of conditional statements, and wherein each of the set of directed graph edges associates a pair of the set of vertices and a direction from a tail vertex of the pair to a head vertex of the pair, a set of statuses, wherein each of the set of statuses is associated with one of the set of vertices, and a set of vertex categories, wherein each of the set of vertex categories is a category value and is associated with a respective vertex of the set of vertices and is determined based a respective conditional statement of the respective vertex, a set of scores, wherein each respective score of the set of scores is associated with a respective vertex and is based a respective conditional statement of the respective vertex; updating the program state data based on a set of inputs comprising a first input, wherein updating the program state data comprises: modifying a status of a first vertex of the set of vertices based on the first input, updating a vertex adjacent to the first vertex; and determining an outcome score based on the set of scores after updating the program state data.
B-21. A tangible, non-transitory, machine-readable medium storing instructions that when executed by one or more processors effectuate operations comprising: obtaining, with one or more processors, identifiers of a plurality of entities; obtaining, with one or more processors, a plurality of symbolic artificial intelligence (AI) models, wherein: each of the plurality of symbolic AI models is configured to produce outputs responsive to inputs based on events caused by at least one of the plurality of entities, at least some of the plurality of entities are associated with outputs of respective symbolic AI models, and at least some of the plurality of entities have respective scores corresponding to the respective outputs of the symbolic AI models; obtaining, with one or more processors, a plurality of scenarios, wherein: each scenario comprises simulated inputs corresponding to one or more simulated events, and at least some scenarios comprise a plurality of simulated inputs; determining, with one or more processors, a population of scores of a given entity among the plurality of entities, wherein respective members of the population of scores correspond to respective outputs of the plurality of symbolic AI models, and wherein the respective outputs correspond to respective scenarios among the plurality of scenarios; and storing, with one or more processors, the population of scores in memory.
B-22. The medium of embodiment B-21, wherein at least one of the plurality of symbolic AI models comprises: a set of vertices and a set of directed graph edges, wherein each of the set of vertices comprises a identifier value and is associated with one of a set of conditional statements, and wherein each of the set of directed graph edges associates a pair of the set of vertices and a direction from a tail vertex of the pair to a head vertex of the pair; a set of statuses, wherein each of the set of statuses is associated with one of the set of vertices; a set of vertex categories, wherein each of the set of vertex categories is a category value and is associated with a respective vertex of the set of vertices and is determined based a respective conditional statement of the respective vertex; and a set of scores, wherein each respective score of the set of scores is associated with a respective vertex and is based a respective conditional statement of the respective vertex.
B-23. The medium of any of embodiments B-21 to B-22, wherein obtaining the plurality of scenarios comprises: determining a first simulated input for a first model of the plurality of symbolic AI models based on a multi-iteration score associated with the first model, wherein the first model is in a first state before updating the first model based on the first simulated input; update the first model based on the first simulated input to advance the first model to a second state, wherein the second state is different from the first state; determine a second input, wherein the second input may be selected based on scores associated with each of a set of possible states associated with the first state; update the first model when it is in the second state based on the second input to advance the second model to a third state, wherein the third state is different from the first state and the second state, and wherein the third state satisfies a terminal state criterion, and wherein a terminal state value is associated with the third state; and update the score associated with the first model based on the terminal state value; and determining a scenario of the plurality of scenarios based on the score.
B-24. The medium of embodiment B-23, wherein determining a first set of simulated inputs comprises determining the first set of inputs based on a first term and a second term, wherein the first term is based on a count of simulations executed that started from the first state and the second term is based on a score value associated with the third state.
B-25. The medium of any of embodiments B-21 to B-24, wherein determining the population of scores comprises using a convolutional neural network to determine a respective score based on values in a respective model of the symbolic AI models.
B-26. The medium of any of embodiments B-21 to B-25, the operations further comprising: fuzzifying the population of scores to provide a set of fuzzified inputs, wherein fuzzifying the outputs comprises using a membership function to determine a degree of membership, and wherein the fuzzified inputs comprises the degree of membership; determine a fuzzified outcome score based on the degree of membership using an inference engine, wherein the inference comprises a set of executable rules that may be matched to the fuzzified inputs; and determine a label associated with a smart contract based on the fuzzified outcome score.
B-27. The medium of any of embodiments B-21 to B-26, wherein obtaining the plurality of scenarios comprises: determining a first scenario for a first symbolic AI model of the plurality of AI models based on a first set of weights corresponding to each of a set of categories, wherein the first symbolic AI model comprises a first plurality of the set of categories; and determining a second scenario for a second symbolic AI model of the plurality of AI models based on the first set of weights, wherein the second symbolic AI model comprises a second plurality of the set of categories.
B-28. The medium of any of embodiments B-21 to B-27, wherein determining the simulated input comprises using a decision tree, wherein the decision tree comprises a first tree node and a second tree node, and wherein the first tree node is associated with a first score, and wherein the first tree node is associated with a second score and wherein the operations further comprise: determining whether the first score is greater than a second score; and in response to the first score being greater than the second score, determining the simulated input based on a value associated with the first tree node.
B-29. The medium of any of embodiments B-21 to B-28, the operations further comprising updating a set of parameters of a neural network based on the population of scores, wherein the neural network provides a weighting value associated with a decision to cancel a self-executing protocol.
B-30. The medium of embodiment B-29, wherein determining the population of scores of a given entity among the plurality of entities comprises determining a sum of the scores.
B-31. A method to perform any of the operations of embodiments B-21 to B-30.
B-32. A method to perform any of the operations of embodiments B-1 to B-19.
B-33. A system, comprising: one or more processors; and memory storing instructions that when executed by the processors cause the processors to effectuate operations comprising: the operations of any one of embodiments B-1 to B-19.
B-34. A system, comprising: one or more processors; and memory storing instructions that when executed by the processors cause the processors to effectuate operations comprising: the operations of any one of embodiments B-21 to B-30.
C-1. A tangible, non-transitory, machine-readable medium storing instructions that, when executed by a computing system, effectuate operations comprising: obtaining, with a computing system, program state of a smart contract, wherein the program state encodes a directed graph, the directed graph comprising: a set of vertices, and a set of directed edges connecting respective pairs of vertices among the set of vertices, wherein the program state includes a set of conditional statements and a set of entities; obtaining, with the computing system, a request encoding a set of conditional statement parameters corresponding to an amendment to the smart contract; determining, with the computing system, a first subset of vertices in the directed graph, wherein each respective vertex of the first subset causes a state change of the program state in response to a respective conditional statement associated with the respective vertex being satisfied; selecting, with the computing system, a second subset of the first subset based on the set of conditional statement parameters encoded in the request; determining, with the computing system, a set of selected entities based on the second subset; determining, with the computing system, whether a set of criteria associated with the set of selected entities is satisfied; updating, with the computing system, the set of conditional statement parameters based on the set of conditional statements in response to a determination that the set of criteria associated with the set of selected entities is satisfied; updating, with the computing system, the second subset based on the updated set of conditional statements; and storing, with the computing system, the program state in storage memory after updating the second subset.
C-2. The medium of embodiment C-1, wherein: the vertices are norm vertices; the first subset is a set of active vertices; the second subset is a set of target vertices; the set of entities include parties to the smart contract; and updating the set of conditional statements comprises: determining an affected conditional statement based on the request, wherein the set of target vertices comprises a reference that is associated with the affected conditional statement; setting an indicator associated with the affected conditional statement to indicate that the affected conditional statement is deprecated; generating a new conditional statement based on the set of conditional statement parameters; and setting the reference to be associated with the new conditional statement.
C-3. The medium of any of embodiments C-1 to C-2, wherein: the directed graph of the program state is stored in persistent memory and comprises a first serialized array of vertices and a second serialized array of vertices; a target vertex of the set of target vertices is a vertex of the first serialized array of vertices and not a vertex of the second serialized array of vertices; and updating the set of target vertices comprises: deserializing the first serialized array of vertices to generate a first deserialized directed graph in a non-persistent memory, wherein the first serialized array of vertices comprises an identifier of the target vertex, and wherein the first deserialized directed graph comprises an adjacency matrix or an adjacency list, and wherein the second serialized array of vertices is not concurrently deserialized, and serializing the first deserialized directed graph in the non-persistent memory to determine an updated first serialized array of vertices, and storing the updated first serialized array of vertices in the persistent memory.
C-4. The medium of any of embodiments C-1 to C-3, wherein the request comprises an entity identifier, and wherein the operations further comprises adding the entity identifier to a set of entities associated with the program state.
C-5. The medium of any of embodiments C-1 to C-4, wherein: the request comprises a first entity identifier associated with a first entity; a first conditional statement of the set of conditional statements comprises a condition that a second entity allocate a resource to a third entity, wherein a second entity identifier is associated with the second entity; and the operations further comprises: determining an entity field used by the first conditional statement, wherein the entity field comprises the second entity identifier, and updating the entity field to comprise the first entity identifier.
C-6. The medium of any of embodiments C-1 to C-5, wherein updating the set of target vertices comprises deleting a first directed edge from the set of directed edges, wherein the first directed edge associates a target vertex of the set of target vertices with a second vertex of the directed graph.
C-7. The medium of any of embodiments C-1 to C-6, the operations further comprising: obtaining a set of confirmation messages from each entity of the set of selected entities, wherein the set of confirmation messages comprises a set of passkey values, and wherein each respective passkey value of the set of passkey values is associated with a respective entity of the set of selected entities; and wherein determining that the set of criteria associated with the set of selected entities is satisfied comprises determining that each respective passkey value of the set of passkey values matches with a respective stored passkey value of a set of stored passkey values.
C-8. The medium of any of embodiments C-1 to C-7, wherein the operations further comprise: determining a first graph structure based on the request, wherein the first graph structure comprises the set of directed edges and a set of logical categories associated with each vertex of the first graph structure; determining whether the first graph structure is identical to a second graph structure of the directed graph; and in response to a determination that the first graph structure is different from the second graph structure with respect to a number of vertices or number of edges, adding a new vertex to the set of vertices based on the request in persistent memory.
C-9. The medium of any of embodiments C-1 to C-8, the operations further comprising: determining a simulated modified program state, wherein determining the simulated modified program state comprises updating a version of the set of conditional statements, and wherein the simulated modified program state comprises the version of the set of conditional statements; determining a set of simulated events, wherein the set of simulated events are determined occur in sequence based on an associated set of occurrence times; and determining a set of outcome program states based on the simulated modified program state and the set of simulated events.
C-10. The medium of any of embodiments C-1 to C-9, wherein the request is a first request of a plurality of requests, the operations further comprising: determining a set of simulated events; determining a set of outcome scores, wherein determining the set of outcome scores comprises, for each respective request of the plurality of requests: determining a respective simulated program state, wherein determining the respective simulated program state comprises updating a version of the set of conditional statements of the program state, and wherein the respective simulated program state comprises the version of the set of conditional statements, determining a respective set of outcome program states based on the respective simulated program state and the set of simulated events, and determining a respective set of outcome scores based on the respective set of outcome program states, wherein the respective set of outcome scores are part of the set of outcome scores; and selecting the first request from the plurality of requests based on the set of outcome scores.
C-11. The medium of any of embodiments C-1 to C-10, wherein the request comprises data encoding a second directed graph, and wherein updating the set of target vertices comprises updating a target vertex of the set of target vertices based on the second directed graph, and wherein an event that triggers a condition of the target vertex causes a second vertex of the second directed graph be set as triggerable.
C-12. The medium of any of embodiments C-1 to C-11, wherein the request comprises an identifier associated with a second directed graph of a second smart contract program, and wherein a target vertex of the set of target vertices is associated with a first priority category value, and wherein the operations further comprise: updating a set of directed edges based on a edge associating the target vertex with a second vertex of the second directed graph; and assigning a second priority category value to the second vertex, wherein: the first priority category value is different from the second priority category value, a first event triggers both the target vertex and the second vertex, and the order by which the target vertex and the second vertex is triggered in response to the first event is determined based on the first priority category value and the second priority category value.
C-13. The medium of any of embodiments C-1 to C-12, wherein: a target vertex of the set of target vertices is associated with a first priority category value; the request comprises a second priority category value: and updating the target vertex comprises changing the first priority category value to the second priority category value.
C-14. The medium of any of embodiments C-1 to C-13, wherein the request is a first request, and wherein the operations further comprise: determining that a previous request failed to satisfy the set of criteria associated with the set of selected entities is satisfied; and determining the first request based on the previous request by modifying a conditional statement parameter encoded in the previous request.
C-15. The medium of any of embodiments C-1 to C-14, wherein the request comprises a set of entity identifiers; determining whether the set of criteria is satisfied comprises determining whether the set of entity identifiers comprises any entities of set of prohibited entities or entities of a set of prohibited entity types.
C-16. The medium of any of embodiments C-1 to C-15, the operations further comprising: obtaining a previous request; determining a simulated modified program state, wherein determining the simulated modified program state comprises updating a version of the set of conditional statements based on the previous request, and wherein the simulated modified program state comprises the version of the set of conditional statements; determining a simulated event; determining that the simulated event causes an allocation of a non-duplicable asset from a first entity to a second entity based on the simulated event and an allocation of the non-duplicable asset from the first entity to a third entity based on the simulated event; and generating a message indicating that the previous request may cause a contradiction in response to a determination that the simulated event causes the allocation of the non-duplicable asset from the first entity to the second entity and the third entity based on the simulated event.
C-17. The medium of any of embodiments C-1 to C-16, wherein determining whether the set of criteria associated with the set of selected entities is satisfied the operations further comprising: storing the set of criteria of the set of selected entities at a storage memory accessible to a first computing device of a distributed computing platform; determining whether the set of criteria associated with the set of selected entities is satisfied using the first computing device; and in response to a determination that the set of criteria associated with the set of selected entities is satisfied using the first computing device, modifying a version of the program state based on the request using a second computing device of the distributed computing platform.
C-18. The medium of any of embodiments C-1 to C-17, wherein: updating the set of conditional statements comprises adding a new conditional statement to the set of conditional statements, wherein the new conditional statement is indexed by a conditional statement identifier; and updating the set of target vertices comprises associating a target vertex of the set of target vertices with the conditional statement identifier.
C-19. The medium of any of embodiments C-1 to C-18, wherein the request comprises a first amount, the operations further comprise: determining whether a target vertex of the set of target vertices is associated with a conditional statement that was triggered by a past event, wherein an outcome of the conditional statement is associated with a transaction between a first entity and a second entity, and wherein the transaction is associated with a score equal to a second amount; determining a difference between the first amount and the second amount; and initiating a transaction between the first entity and the second entity based on the difference.
C-20. The medium of any of embodiments C-1 to C-19, the operations comprising steps for amending the smart contract while the smart contract is active.
C-21. The medium of any of embodiments C-1 to C-20, the operations comprising: steps for amending the smart contract in a manner that affects only a subset of parties to the smart contract.
C-22. A system comprising: one or more processors; and memory storing instructions that, when executed by at least one of the one or more processors, causes at least one of the one or more processors to effectuate any of the operations of embodiments C-1 to C-21.
C-23. A method to perform the operations of any of the embodiments C-1 to C-21.
D-1. A tangible, non-transitory, machine-readable medium storing instructions that when executed by one or more processors effectuate operations comprising: obtaining, with a computer system, a first directed graph of a first program state of a symbolic artificial intelligence (AI) model, wherein: the first directed graph comprises a first set of vertices and a set of directed edges, the first directed graph encodes a set of conditions and is associated with a first entity and a second entity, the conditions being conditional statements, each respective vertex of the first set of vertices is associated with a status among a set of types of status, wherein the set of types of status comprises a first status indicating that the respective vertex is satisfied, a second status indicating that the respective vertex is failed, and a third status indicating that the respective vertex is not satisfied but satisfiable, and each respective vertex of the first set of vertices is categorized in a vertex category among a set of vertex categories, the set of vertex categories comprising a first category, and the first set of vertices comprises a first vertex and a second vertex, wherein the first vertex is categorized as being of the first category, and wherein: a directed edge associates with first vertex with the second vertex, and the first vertex is associated with a first condition satisfiable by a first event caused by the first entity, and wherein a change of status of the first vertex from the third status causes the status of the second vertex to be changed to the third status; simulating, with the computer system, evolving program state of the symbolic AI model from the first program state by evaluating conditions of the set of conditions of the first directed graph to form a second directed graph, wherein: the second directed graph comprises a second set of vertices, and wherein a third vertex of the second set of vertices is associated with a second condition satisfiable by a second event caused by the second entity, simulating comprises determining a set of action values of the first entity based on the second directed graph by: determining a set of reward values based on a second set of conditions associated with the second directed graph, wherein each of the set of reward values is associated with a vertex of the second directed graph, and determining the set of action values based on the set of reward values and a set of paths starting from the first vertex to a terminal vertex; and determining and storing in memory, with the computer system, an outcome program state based on the set of action values, wherein the outcome program state is different from the first program state.
D-2. The medium of embodiment D-1, wherein evolving from the first program state further comprises: determining a set of active vertices of the first set of vertices, wherein each respective active vertex of the set of active vertices may be triggered by an action of the first entity to triggers an adjacent vertex of the respective active vertex; and determining a first set of child vertices, wherein each respective child vertex of the first set of child vertices is adjacent to one respective active vertex of the set of active vertices, wherein the second set of vertices comprises the first set of child vertices.
D-3. The medium of any of embodiments D-1 to D-2, wherein evolving from the first program state further comprises evolving from the first program state to each of a set of terminal program states over a plurality of simulated state evolutions, wherein each respective simulated state evolution determines a respective path through the second directed graph that ends at a terminal program state, and wherein the plurality of simulated state evolutions provide a plurality of paths such that each terminal vertex of the second directed graph is in at least one of the plurality of paths.
D-4. The medium of any of embodiments D-1 to D-3, wherein determining the set of action values further comprises: determining an initial set of actions using a trained neural network based on the first directed graph, wherein each respective action of the initial set of actions is performable by the first entity and is associated with a respective score of an initial set of action values determined by the trained neural network; and iteratively traversing the second directed graph based on the initial set of actions and the initial set of action values using a tree search operation to determine the set of action values.
D-5. The medium of embodiment D-4, wherein iteratively traversing the second directed graph comprises: determining a first heuristic value based on the first category, wherein the first heuristic value is associated with satisfaction of the first vertex; determining a second heuristic value based on the first category, wherein the second heuristic value is associated with failure of the first vertex; and determining whether the first vertex is satisfied based on the first heuristic value and the second heuristic value, wherein the first heuristic value is associated with a greater probability of selection than the second heuristic value.
D-6. The medium of any of embodiments D-1 to D-5, wherein the outcome program state is a first predicted outcome program state, the operations further comprise: determining a set of outcome program states comprising the first predicted outcome program state; obtaining an event performed by the second entity, wherein the second event causes the program state to change to an actual outcome program state; determining whether an unexpected event threshold is satisfied based on whether the actual outcome program state is not in the set of outcome program states; and sending a message indicating that the unexpected event threshold is satisfied.
D-7. The medium of any of embodiments D-1 to D-6, wherein the operations further comprise: determining a transaction graph based on a set of smart contract programs, wherein a first transaction graph vertex of the transaction graph is associated with the first entity, and wherein a second transaction graph vertex of the transaction graph is associated with the second entity; determining a transaction path between the first entity and the second entity; determining an inter-entity score based on the transaction path; and wherein determining the outcome program state comprises determining the outcome program state based on the inter-entity score.
D-8. The medium of embodiment D-7, wherein the operations further comprise: determining that the transaction path is a cyclical path, wherein the cyclical path comprises a set of transaction graph edges that are connected and begin and end at a same transaction graph vertex; and storing a value indicating that the transaction path is cyclical.
D-9. The medium of any of embodiments D-7 to D-8, wherein determining the transaction graph comprises: traversing a set of directed graphs of the set of smart contract programs to determine a set of score changes between a pair of entities of the transaction graph; and updating, for each of the set of score changes between the pair of entities, a transaction graph edge associating the pair of entities.
D-10. The medium of any of embodiments D-7 to D-9, wherein the operations further comprise: determining a set of smart contract programs associated with the first transaction graph vertex, wherein each smart contract program of the set of smart contract programs is determined to cause a score change for the first entity based on conditions of the set of smart contract programs; and determining a set of contribution weights, wherein each respective contribution weight of the set of contribution weights is associated with a respective smart contract program of the set of smart contract programs, and wherein each respective contribution weight is correlated with a ratio by which the respective smart contract program contributes to a net score change of the first entity.
D-11. The medium of any of embodiments D-7 to D-10, wherein determining the transaction path between the first entity and the second entity the operations further comprise: obtaining a transaction path threshold; determining whether a set of transaction graph edges of the transaction graph satisfies the transaction path threshold; and responsive to a determination that the set of transaction graph edges satisfies the transaction path threshold, determining the transaction path between the first entity and the second entity based on the set of transaction graph edges.
D-12. The medium of any of embodiments D-1 to D-11, and wherein satisfying a failure threshold of the vertex causes the vertex to be associated with the second status, and wherein the operations further comprise: determining whether the second entity is associated with a third event caused by the second entity that resulted in a fourth vertex being associated with the second status, wherein the fourth vertex is associated with the first category; and responsive to the second entity being associated with the third event caused by the second entity that resulted in the fourth vertex being associated with the second status, reduce a reward value determined from a score transfer from the second entity to the first entity.
D-13. The medium of any of embodiments D-1 to D-12, wherein the operations further comprise categorizing one or more quantitative values of the first program state before determining the set of action values.
D-14. The medium of any of embodiments D-1 to D-13, the operations further comprising determining whether the first vertex should be associated with the second status based on determining whether a time point satisfies a failure time threshold.
D-15. The medium of any of embodiments D-1 to D-14, wherein determining the set of action values of the first entity comprises using an intelligent agent, wherein the intelligent agent comprises: a set of stored parameters; a first routine to update the set of stored parameters one or more times; a second routine to determine action values based on the set of stored parameters and on the first program state.
D-16. The medium of embodiment 15, wherein using the intelligent agent comprises: determining a first path score associated with a first path from the first vertex to a terminal vertex of the second directed graph; determining a first weight based on reaching the first program state from an initial program state; determining an intermediate program state based on an event that changes a status of the first vertex to the first status or the second status; determining a second weight based on reaching a terminal program state from the intermediate program state; determining a counterfactual regret value based on a summation comprising a product of the first path score, the first weight, and the second weight; and determining the set of action values based on the counterfactual regret value.
D-17. The medium of any of embodiments D-1 to D-16, wherein determining the set of reward values based on the second set of conditions comprises: determining a threshold value to satisfy the first vertex based on a condition associated with the first vertex; and determining a first reward value based on the threshold value, wherein the first reward value is associated with the first vertex.
D-18. The medium of any of embodiments D-1 to D-17, wherein obtaining the first directed graph of the first program state further comprises: determining whether the first program state is different from a previous program state; and responsive to a determination that the first program state is different from the previous program state, obtaining the first directed graph of the first program state.
D-19. The medium of any of embodiments D-1 to D-18, wherein determining the second directed graph comprises: varying a first set of threshold values of the set of conditions to determine a set of modified values; obtaining an initial set of events causable by the first entity based on the set of modified values; determining a set of possible events based on the initial set of events causable by the first entity using a first trained neural network, wherein the first trained neural network is trained using a set of self-play operations; and determining the second directed graph based on the set of possible events.
D-20. The medium of any of embodiments D-1 to D-19, wherein evolving from the first program state comprises: obtaining a failure penalty; wherein the second directed graph comprises a first path that includes a vertex having the second status; and modifying an action value associated with the first path based on the failure penalty.
D-21. The medium of any of embodiments D-1 to D-20, wherein the operations further comprise: determining an outcome score associated with the outcome program state; determining whether the outcome score satisfies an outcome score threshold; and responsive to the outcome score satisfying the outcome score threshold, generate an alert indicating that the outcome score threshold is satisfied.
D-22. The medium of any of embodiments D-1 to D-21, wherein determining the action value further comprises determining an action value based on an entity property associated with the first entity or the second entity.
D-23. The medium of any of embodiments D-1 to D-22, the operations further comprising determining the set of paths by performing a set of simulated state evolutions, wherein performing each respective simulated state evolution of the set of simulated state evolutions comprises: determining a set of computed values using a random, pseudorandom, or quasi-random number generation method; and selecting a respective subset of vertices of a respective path based on the set of computed values.
D-24. The medium of embodiment 23, wherein a selectable vertex in the second directed graph is not in any of the subsets of vertices determined from the set of simulated state evolutions.
D-25. The medium of any of embodiments D-1 to D-24, wherein simulating program state evolution comprises simulating program state evolution for at least 10,000 iterations from an initial program state to a terminal program state.
D-26. The medium of any of embodiments D-1 to D-25, wherein a count of vertices in the first directed graph is greater than 1000 vertices.
D-27. The medium of any of embodiments D-1 to D-26, the operations further comprising a means of determining the outcome program state.
D-28. The medium of any of embodiments D-1 to D-27, wherein the operations further comprise: determining a transaction graph based on a set of smart contract programs, wherein a first transaction graph vertex of the transaction graph is associated with the first entity, and wherein a second transaction graph vertex of the transaction graph is associated with the second entity; determining a transaction path between the first entity and the second entity; determining an inter-entity score based on the transaction path; and wherein determining the outcome program state comprises determining the outcome program state based on the inter-entity score.
D-29. The medium of any of embodiments D-1 to D-28, wherein the operations further comprise: determining a plurality of outcome states; and determining a population score based on the plurality of outcome states.
D-30. A tangible, non-transitory, machine-readable medium storing instructions that when executed by one or more processors effectuate operations comprising: obtaining, with one or more processors, a symbolic artificial intelligence (AI) model, wherein the symbolic AI model is configured to produce an outcome state responsive to an input based on events, wherein at least some of the events are caused by a first entity or a second entity; obtaining, with one or more processors, a first scenario and a second scenario, wherein the first scenario causes the failure of a condition associated with a norm of the symbolic AI model, and wherein the second scenario satisfies the condition associated with the norm of the symbolic AI model; obtaining, with one or more processors, a failure penalty value; determining, with one or more processors, a first outcome state based on the symbolic AI model, the first scenario, and the failure penalty value; determining, with one or more processors, a second outcome state based on the symbolic AI model and the second scenario; determining, with one or more processors, an outcome score based on the first outcome state and the second outcome state; and storing, with one or more processors, the outcome score in memory.
D-31. A method to perform the operations of any of the embodiments D-1 to D-30.
D-32. A system comprising: one or more processors; and memory storing instructions that, when executed by at least one of the one or more processors, causes at least one of the one or more processors to effectuate any of the operations of embodiments D-1 to D-30.
E-1. A tangible, non-transitory, machine-readable medium storing instructions that, when executed by a computing system, effectuate operations comprising: obtaining, with a computing system, program state of a self-executing protocol, wherein the program state encodes: a set of conditional statements; a set of entities, wherein the set of entities comprises a first entity; a directed graph, the directed graph comprising: a set of vertices, wherein each respective vertex of the set of vertices is associated with a respective category label of a set of mutually exclusive categories; a set of directed edges connecting respective pairs of vertices among the set of vertices; obtaining, with the computing system, an entity profile of the first entity, wherein: the entity profile comprises a first graph portion template, the first graph portion template comprises a first vertex template and an edge template, the first vertex template is associated in memory with a first category label of the set of mutually exclusive category labels, and the edge template specifies an edge direction to or from a vertex matching the first vertex template; determining, with the computing system, whether the first graph portion template matches a graph portion in the directed graph based on a first vertex of the directed graph matching the first vertex template and a first directed edge of the directed graph matching the edge template; determining, with the computing system, an outcome score based on the first graph portion template matching the graph portion in the directed graph; determining, with the computing system, whether the outcome score satisfies an outcome score threshold; and in response to the outcome score satisfying the outcome score threshold, storing, with the computing system, a value indicating that the outcome score satisfies the outcome score threshold.
E-2. The medium of embodiment E-1, wherein: the set of vertices are a set of norm vertices; the first vertex is a first norm vertex; the set of entities include parties to the self-executing protocol; the operations further comprising: obtaining a plurality of self-executing protocol programs comprising a plurality of directed graphs, wherein each respective directed graph of the plurality of directed graphs is associated with a respective set of entities that comprises the first entity; determining the first graph portion template based on the plurality of directed graphs, wherein a second norm vertex of the plurality of directed graphs matches the first norm vertex template of the first graph portion template, and wherein a condition of the second norm vertex is indicated to have been failed by the first entity based on an event message; and determining an outcome determination parameter based on a number of times that the first graph portion template matches with a respective graph portion in the plurality of self-executing protocol programs, wherein determining the outcome score comprises determining the outcome score based on the outcome determination parameter.
E-3. The medium of any of embodiments E-1 to E-2, the operations further comprising: obtaining a plurality of self-executing protocol programs comprising a plurality of directed graphs, wherein each respective self-executing protocol program of the plurality of self-executing protocol programs comprises a respective directed graph of the plurality of directed graphs; determining the first graph portion template based on the plurality of self-executing protocol programs, wherein a second vertex of a second directed graph of the plurality of directed graphs matches the first vertex template, and wherein a third vertex of the plurality of directed graphs matches a second vertex template, and wherein a condition of the third vertex is indicated as having been satisfied based on an event message; and determining an outcome determination parameter based on a number of times that the first graph portion template matches with a respective graph portion in the plurality of self-executing protocol programs, wherein determining the outcome score comprises determining the outcome score based on the outcome determination parameter.
E-4. The medium of any of embodiments E-1 to E-3, wherein the entity profile is a first entity profile, and wherein the operations further comprise: determining a transaction score based on the directed graph, wherein the transaction score is associated with a transaction between the first entity and a second entity; and updating an association between the first entity profile and a second entity profile based on the transaction score, wherein the second entity profile is associated with the second entity.
E-5. The medium of any of embodiments E-1 to E-4, the operations further comprising: determining whether the first entity has failed a conditional statement associated with a second vertex of the directed graph; and in response to a determination that the first entity has failed the conditional statement, updating an entity score of an entity graph, wherein the entity score is associated with the first entity, and wherein the entity graph comprises a plurality of entity vertices, and wherein each respective entity vertex of the plurality of entity vertices is associated with a respective entity profile.
E-6. The medium of embodiment E-5, wherein the entity graph is stored on a distributed, tamper-evident ledger, and wherein updating the entity score comprises: obtaining an encryption key associated with the first entity; obtaining a previous entity score from the distributed, tamper-evident ledger based on the encryption key; and updating the entity score based on the previous entity score.
E-7. The medium of any of embodiments E-5 to E-6, wherein the entity graph is stored on a distributed, tamper-evident ledger, and wherein the operations further comprise: determining whether the entity score satisfies an entity score threshold of a verification entity; and in response to the entity score satisfying the entity score threshold, storing an indicator that the first entity satisfies the entity score threshold of the verification entity.
E-8. The medium of any of embodiments E-5 to E-7, the operations further comprising: determining whether the entity score satisfies an entity score threshold of a verification entity; and sending a message to an application program interface, wherein the message indicates that the first entity satisfies the entity score threshold of the verification entity.
E-9. The medium of any of embodiments E-5 to E-8, wherein the entity profile is a first entity profile, and wherein the operations further comprise: determining a second entity score associated with the first entity, wherein the first entity profile does not comprise the second entity score; obtaining a passkey value; and in response to receiving the passkey value, sending a message comprising the second entity score.
E-10. The medium of any of embodiments E-1 to E-9, wherein determining the outcome score comprises determining the outcome score using a neural network based on a feature set, wherein: determining the feature set, wherein determining the feature set comprises determining whether the first graph portion template matches a graph portion in the directed graph; and the neural network is trained on a plurality of directed graphs of a plurality of a self-executing protocol programs, wherein the first graph portion template matches a graph portion of a subset of the plurality of directed graphs.
E-11. The medium of any of embodiments E-1 to E-10, wherein determining the outcome score comprises: generating a set of embeddings based on a set of vertices of the directed graph, wherein each vertex of the set of vertices is associated with an embedding of the set of embeddings, and wherein each embedding comprises a vector; determining a feature set based on the set of embeddings; and determining the outcome score using a neural network based on the feature set.
E-12. The medium of any of embodiments E-1 to E-11, wherein the entity profile is a first entity profile and the outcome score is a first outcome score, and wherein the operations further comprise: obtaining a second entity profile, wherein the second entity profile is associated with a second entity, and wherein the second entity profile comprises the first graph portion template, and wherein a second outcome determination parameter is determined based on the first graph portion template; determining a second outcome score associated with the second entity profile based on the second outcome determination parameter; and selecting the first entity based on the first outcome score and the second outcome score.
E-13. The medium of any of embodiments E-1 to E-12, the operations further comprising: sampling the directed graph to determine a set of subgraphs; determining a vector based on the set of subgraphs using a skip-gram model; and determining the outcome score using a neural network based on the vector.
E-14. The medium of any of embodiments E-1 to E-13, wherein the first graph portion template further comprises a second vertex template, wherein the second vertex template is associated with a second category label of the set of mutually exclusive category labels, and wherein the second category label is different from the first category label.
E-15. The medium of any of embodiments E-1 to E-14, the operations further comprising: updating the entity profile based a history of the first entity; storing the entity profile on a centralized computing platform, wherein the entity profile is associated with an entity identifier; and updating a value associated with the entity identifier, wherein the value is stored on a distributed, tamper-evident ledger operating on a distributed computing platform.
E-16. The medium of any of embodiments E-1 to E-15, wherein the entity profile is a first entity profile, and wherein the operations further comprising: obtaining a second entity profile; determining whether a set of entity similarity criteria is satisfied based on the first entity profile and the second entity profile; and storing value indicating that the first entity profile and the second entity profile satisfy the set of entity similarity criteria.
E-17. The medium of any of embodiments E-1 to E-16, wherein the first graph portion template further comprises a second vertex template, wherein the second vertex template is not connected to the first vertex template in the first graph portion template by any edge templates.
E-18. The medium of any of embodiments E-1 to E-17, wherein the directed graph is a first self-executing protocol directed graph, and wherein the operations further comprise: determining a first transaction amount between the first entity and a second entity based on the first self-executing protocol directed graph; determining a second transaction amount between the second entity and a third entity based on a second self-executing protocol directed graph; updating a first association between the first entity and the second entity of an entity graph based on the first transaction amount; updating a second association between the second entity and the third entity of the entity graph based on the second transaction amount; and determining whether the first entity is associated with the third entity based on the first association, the first transaction amount, the second association, and the second transaction amount.
E-19. The medium of embodiment 18, the operations further comprising: determining whether the first entity has failed a conditional statement associated with the first vertex; in response to a determination that the first entity has failed the conditional statement, updating an entity score is associated with the first entity; and sending a message to the third entity in response to the updating of the entity score associated with the first entity.
E-20. A method to perform the operations of any of the embodiments E-1 to E-19.
E-21. A system comprising: one or more processors; and memory storing instructions that, when executed by at least one of the one or more processors, causes at least one of the one or more processors to effectuate any of the operations of embodiments E-1 to E-19.
F-1. A tangible, non-transitory, machine-readable medium storing instructions that, when executed by a computing system, effectuate operations comprising: obtaining, with a computing system, a set of conditions; obtaining, with the computing system, a first cross-program entity identifier of a first entity, wherein the first cross-program entity identifier is unique amongst a set of cross-program entity identifiers of a decentralized computing platform; obtaining, with the computing system, a set of directed graphs of a set of self-executing protocols comprising a first self-executing protocol and a second self-executing protocol that are executed on the decentralized computing platform, wherein: each respective self-executing protocol of the set of self-executing protocols comprises data of a respective directed graph of the respective self-executing protocol, and the first cross-program entity identifier is associated with a first program-specific entity identifier of the first self-executing protocol and a second program-specific entity identifier of the second self-executing protocol; determining, with the computing system, that the set of conditions is applicable to the first entity based on the first cross-program entity identifier; determining, with the computing system, whether the set of conditions are satisfied based on whether a graph portion associated with the set of directed graphs corresponds to a graph portion template of the set of conditions; and in response to a determination that the graph portion corresponds to the graph portion template, storing, with the computing system, an indication that the first entity violated the set of conditions in a profile of the first entity using the first cross-program entity identifier.
F-2. The medium of embodiment F-1, the operations further comprising: determining a first set of geographic locations associated with the first entity based on the first cross-program entity identifier; and determining whether the first set of geographic locations satisfies a first condition of the set of conditions based on whether the first set of geographic locations is within a geofence indicated by the first condition, wherein the indication indicates that the first entity violated the set of conditions based on whether the first set of geographic locations satisfies the first condition.
F-3. The medium of any of embodiments F-1 to F-2, the operations further comprising determining a second set of counterparty entities based on the set of self-executing protocols, wherein each counterparty entity of the set of counterparty entities is associated with a transaction with the first entity.
F-4. The medium of any of embodiments F-1 to F-3, wherein obtaining the set of conditions comprises: obtaining a governing document; determining a set of entity categories using a natural language processing model based on governing document; and determining a condition of the set of conditions based on the set of entity categories.
F-5. The medium of any of embodiments F-1 to F-4, the operations further comprising: obtaining a governing document; selecting a section of the governing document based on a text header indicated by a set of text sizes or text spacings; and determining a condition of the set of conditions based on the section of the governing document.
F-6. The medium of any of embodiments F-1 to F-5, the operations further comprising: obtaining a first profile associated with the first cross-program entity identifier; obtaining a natural language document, wherein the natural language document comprises a verifying agent identifier and an entity name associated with the first cross-program entity identifier; using a natural language processing model to parse the natural language document to determine the verifying agent identifier and the entity name; sending a first message comprising the entity name to an application program interface (API) of a third-party entity based on the verifying agent identifier; and obtaining a second message from the third-party entity indicating that the entity name is valid and, in response, setting the first profile associated with the first cross-program entity identifier as a verified profile.
F-7. The medium of any of embodiments F-1 to F-6, the operations further comprising sending a notification message to a second entity indicating that the first entity failed the set of conditions.
F-8. The medium of any of embodiments F-1 to F-7, the operations further comprising: sending a first message comprising data of a pending transaction to a third entity, wherein a participant of the pending transaction is associated with the first cross-program entity identifier; obtaining a second message from the third entity, wherein the second message indicates that the third entity has verified the pending transaction; and in response to receiving the second message, storing a value indicating that the transaction was verified by the third entity on a distributed, tamper-evident data structure.
F-9. The medium of any of embodiments F-1 to F-8, the operations further comprising: determining, after a threshold duration of time after determining whether the set of conditions are satisfied, whether the set of conditions are satisfied a second time; and in response to a determination that the set of conditions are satisfied, setting a value to indicate that a resource transfer or allocation of a pending transaction is permitted, wherein a participant of the pending transaction is associated with the first cross-program entity identifier.
F-10. The medium of any of embodiments F-1 to F-9, the operations further comprising: determining that a variable of the set of conditions is not stored in data of a smart self-executing protocol; compute a value for the variable using a function encoded in the set of conditions; determining whether a the value satisfies a threshold value of a first condition; and in response to a determination that the value satisfies the threshold value, storing a value indicating that the first entity satisfies the first condition to a persistent storage.
F-11. The medium of any of embodiments F-1 to F-10, the operations further comprising: obtaining an additional governing document; updating the set of conditions based on the additional governing document; and determining whether the updated set of conditions is satisfied.
F-12. The medium of any of embodiments F-1 to F-11, wherein determining whether the set of conditions is satisfied further comprises: determining a first score change of the first self-executing protocol; determining that the first score change is associated with the first entity based on an association between the first program-specific entity identifier and the first cross-program entity identifier; determining a second score change of the second self-executing protocol; determining that the second score change is associated with the first entity based on an association between the second program-specific entity identifier and the first cross-program entity identifier; and determining whether the first entity satisfies the set of conditions based on the first score change and the second score change.
F-13. The medium of any of embodiments F-1 to F-12, the operations further comprising: determine a summation based on the first score change and the second score change, wherein determining whether the set of conditions is satisfied comprises determining whether the summation satisfies a threshold value.
F-14. The medium of any of embodiments F-1 to F-13, wherein a set of entities participating in the first self-executing protocol do not have permission to view the first cross-program entity identifier and the computer system prevents such viewing responsive to the lack of permission.
F-15. The medium of any of embodiments F-1 to F-14, the operations further comprising: determining whether a first value of a transaction satisfies a warning threshold, wherein the warning threshold is based on a condition of the set of conditions; and sending a message indicating that the warning threshold has been satisfied to the first entity.
F-16. The medium of any of embodiments F-1 to F-15, the operations further comprising: determining a hierarchy of conditions based on a set of precedence values associated with the set of conditions; determining a pair of conflicting conditions based on the set of conditions and a difference in labels between category labels of the set of conditions, wherein each category label of a respective condition of the set of conditions is one of a set of mutually exclusive category labels; and determining an overriding condition based on the hierarchy of governing conditions, wherein the overriding condition is one of the pair of conflicting conditions, and wherein the overriding condition is indicated to take precedence over the other condition of the pair of conflicting conditions.
F-17. The medium of any of embodiments F-1 to F-16, the operations further comprising: determining that a second cross-program entity identifier is associated with the first entity; determining that a condition is associated with the second cross-program entity identifier; generating an association between the first cross-program entity identifier and the second cross-program entity identifier in a database of cross-program entity identifiers; and persisting the database of cross-program entity identifiers to a persistent storage of the computing system.
F-18. The medium of embodiment F-17, the operations further comprising steps for obtaining the set of conditions.
F-19. The medium of any of embodiments F-1 to F-18, the operations further comprising steps for determining whether the set of conditions is violated.
F-20. A method to perform the operations of any of the embodiments F-1 to F-19.
F-21. A system, comprising: one or more processors; and memory storing instructions that when executed by the processors cause the processors to effectuate operations comprising: the operations of any one of embodiments F-1 to F-19.
G-1. A tangible, non-transitory, machine-readable medium storing instructions that, when executed by a computing system, effectuate operations comprising: executing, with one or more processors, an instance of an application, wherein: program state of the instance of the application comprises a set of vertices and a set of directed graph edges, each vertex of the set of vertices comprises an identifier and a category label of a set of mutually exclusive category labels, and each of the set of directed graph edges associates a pair of the set of vertices and a direction from a tail vertex of the pair to a head vertex of the pair; serializing, with one or more processors, the set of vertices in a serialized data format and storing a record comprising the serialized set of vertices in a first persistent storage of a first computing device of a plurality of computing devices communicatively coupled to each other via a network; distributing, with one or more processors, the serialized set of vertices to a second computing device of the plurality of computing devices; deserializing, with one or more processors, the serialized set of vertices with the second computing device to generate a second instance of a directed graph, wherein the second instance of the directed graph encodes the set of vertices and the set of directed graph edges in non-serialized data format; storing, with one or more processors, the second instance of the directed graph in a second persistent storage, wherein the second persistent storage is a local persistent storage of the second computing device; receiving, with one or more processors, a message encoding a graph portion template, wherein the graph portion template comprises a first vertex template and an edge template, and wherein the first vertex template is associated with a first category label of the set of mutually exclusive category labels, and wherein the edge template specifies an edge direction to or from a vertex matching the first vertex template; querying, with one or more processors, a data structure of the second persistent storage based on the graph portion template to retrieve a subset of vertices encoded in the second instance of the directed graph; computing, with one or more processors, a response value based on the subset of vertices; and sending, with one or more processors, a response comprising the response value from the second computing device to a response destination indicated by the message.
G-2. The medium of embodiment G-1, wherein the computing the response value comprises: determining a set of entities based on the subset of vertices; determining whether a data retrieval criterion of the set of entities is satisfied; and wherein computing the response value comprises computing the response value in response to a determination that the data retrieval criterion is satisfied.
G-3. The medium of any of embodiments G-1 to G-2, wherein the message comprises instructions to update the directed graph, the operations comprising: storing an updated directed graph in the second persistent storage based on the message; determining a first update confirmation value associated with the updated directed graph; receiving a second update confirmation value at the second computing device; determining whether the first update confirmation value satisfies a set of storage update criteria based on the second update confirmation value; and in response to a determination that the first update confirmation value satisfies the set of storage update criteria, set an indicator to indicate that the updated directed graph is valid in the second persistent storage.
G-4. The medium of any of embodiments G-1 to G-3, the operations further comprising: determining whether a graph portion of the directed graph matches a graph portion template of a library of graph portion templates; and in response to a determination that the graph portion of the directed graph matches the graph portion template, associate an index value with a first vertex of the graph portion, wherein the first vertex is retrievable with the index value.
G-5. The medium of any of embodiments G-1 to G-4, the operations further comprising: selecting a plurality of directed graphs stored in the second persistent storage based on the message; for each respective directed graph of the plurality of directed graphs, determining a respective subset of vertices associated with the respective directed graph; and wherein determining the response value based on the respective subsets of vertices of the plurality of directed graphs.
G-6. The medium of any of embodiments G-1 to G-5, the operations further comprising: detecting a state-updating event associated with an update to program state, wherein the state-updating event causes a change in the set of vertices or a status associated with the set of vertices; updating a historical sequence of records based on the state-updating event, wherein each respective vertex of the historical sequence of records is associated with a respective state-updating event; and storing the historical sequence of records in the second persistent storage, wherein each vertex of the historical sequence of records is associated with a respective version of the directed graph.
G-7. The medium of any of embodiments G-1 to G-6, wherein the graph portion template comprises a plurality of vertex templates and a plurality of directed graph edges associating the plurality of vertex templates with each other.
G-8. The medium of any of embodiments G-1 to G-7, wherein the plurality of computing devices is a first plurality of computing devices, the operations further comprising: selecting a subset of nodes from the plurality of computing devices, where each respective node of the subset of nodes corresponds with a respective computing device of the plurality of computing devices, wherein the subset of nodes is fewer in number than the total number of nodes of the plurality of computing devices; and generating, with at least one the subset of nodes, a block of a sequence of blocks stored in the first persistent storage.
G-9. The medium of any of embodiments G-1 to G-8, wherein the first computing device validates an update to program state based on a consensus voting process involving other computing devices in the plurality of computing devices.
G-10. The medium of any of embodiments G-1 to G-9, wherein the directed graph is a first directed graph, the operation further comprising: determining a predicted directed graph based on the first directed graph; and storing the predicted directed graph in association with the first directed graph.
G-11. The medium of any of embodiments G-1 to G-10, the operations further comprising validating the directed graph at the second computing device based on a signature value provided by the first computing device.
G-12. The medium of any of embodiments G-1 to G-11, the operations further comprising storing a third instance of the directed graph in a third persistent storage, wherein the third persistent storage is a persistent storage of a cloud computing server.
G-13. The medium of any of embodiments G-1 to G-12, wherein the second persistent storage is a persistent storage of a peer-to-peer data-sharing network comprising a second plurality of computing devices, wherein data stored on a first peer of the second plurality of computing devices is distributed to other peers of the second plurality of computing devices.
G-14. The medium of any of embodiments G-1 to G-13, the operations further comprising storing a natural language document in association with the directed graph in the data structure of the second persistent storage.
G-15. The medium of any of embodiments G-1 to G-14, the operations comprising steps for serializing, steps for storing, and steps for deserializing the directed graph.
G-16. The medium of any of embodiments G-1 to G-15, the operations comprising steps for updating program state of the application.
G-17. The medium of any of embodiments G-1 to G-16, the operations further comprising: generating a plurality of previous versions of the directed graph based on a sequence of blocks stored in the first persistent storage; and storing the plurality of previous versions in the second persistent storage.
G-18. The medium of any of embodiments G-1 to G-17, wherein storing the second instance of the directed graph in the second persistent storage further comprises: determining an updated vertex of the second instance of the directed graph; and storing a score change or a set of entities associated with the updated vertex in the data structure of the second persistent storage.
G-19. The medium of any of embodiments G-1 to G-18, wherein computing the response value based on the subset of vertices comprises: determining a set of conditional statements associated with the subset of vertices; determining a set of scores based on the set of conditional statements; and determining the response value based on a sum of the set of scores.
G-20. A method to perform the operations of any of the embodiments G-1 to G-19.
G-21. A system, comprising: one or more processors; and memory storing instructions that when executed by the processors cause the processors to effectuate operations comprising: the operations of any one of embodiments G-1 to G-19.
H-1. A tangible, non-transitory, machine-readable medium storing instructions that, when executed by a computing system, effectuate operations comprising: obtaining, with the computing system, a first set of conditional statements, wherein each of the first set of conditional statements is associated with a vertex of a first directed graph of a distributed application listing a first entity and second entity as associated with the distributed application; determining, with the computing system, a subset of conditional statements from the first set of conditional statements based on a first category selected from a set of mutually exclusive categories, wherein: triggering each respective conditional statement of the subset of conditional statements causes a state associated with the respective conditional statement to be updated from an initial state to a different state; each of the subset of conditional statements is associated with the first category; and each of the subset of conditional statements is indicated to be triggered based on a first event, wherein the first event comprises a value indicating a resource amount; generating, with the computing system, an integrated test condition based on the subset of conditional statements, wherein the integrated test condition is associated with a shared resource type and a numeric value; obtaining, with the computing system, a second directed graph, wherein the second directed graph is associated with the first entity; determining, with the computing system, a simulated event based on a first conditional statement of the second directed graph; determining, with the computing system, whether the simulated event triggers the integrated test condition; and storing, with the computing system, a result indicating that the simulated event triggers the integrated test condition in response to a determination that the simulated event triggers the integrated test condition.
H-2. The medium of embodiment H-1, wherein determining the simulated event comprises simulating adding an additional vertex and directed edge to the first directed graph, wherein the additional vertex is associated with a parent vertex via the additional directed edge, wherein the parent vertex is associated with the first conditional statement.
H-3. The medium of any of embodiments H-1 to H-2, wherein determining the simulated event comprises simulating a sequence of simulated events, wherein each respective simulated event of the sequence of simulated events is indicated to at a different time.
H-4. The medium of any of embodiments H-1 to H-3, the operations further comprising determining that the simulated event triggers the integrated test condition, wherein the triggering of the integrated test condition causes the integrated test condition to indicate a specified conditional statement of the subset of conditional statements as failed.
H-5. The medium of any of embodiments H-1 to H-4, the operations further comprising sending a message to the first entity in response to a determination that the simulated event triggered the integrated test condition, wherein the message indicates that the simulated event triggered the integrated test condition.
H-6. The medium of any of embodiments H-1 to H-5, wherein: the first directed graph are associated with a first smart contract associated with a first set of entities; each of the first set entities is associated with a conditional statement of one or more vertices of the first directed graph; the second directed graph is associated with a second smart contract associated with a second set of entities; each of the second set entities is associated with a conditional statement of one or more vertices of the second directed graph; and the first set of entities is different from the second set of entities.
H-7. The medium of any of embodiments H-1 to H-6, the operations further comprising partitioning a parent directed graph into the first directed graph and the second directed graph.
H-8. The medium of any of embodiments H-1 to H-7, wherein the first directed graph is associated with a first natural language document, the second directed graph is associated with a second natural language document, and the operations further comprise: visually indicating a first text section of the first natural language document, wherein the first text section is associated with a parent vertex, wherein the parent vertex is associated with the conditional statement; and in response to a determination that the event triggers the second conditional statement of the second directed graph, visually indicating a second section of the second natural language document, wherein the first text section is associated with the parent vertex.
H-9. The medium of any of embodiments H-1 to H-8, the operations further comprising retrieving the second directed graph based on data associated with the first entity, wherein the event is caused by the first entity.
H-10. The medium of any of embodiments H-1 to H-9, wherein the first directed graph was previously updated from a previous version of the first directed graph, wherein the updated first directed graph comprises an active vertex that was not active in the previous version of the first directed graph, the operations further comprising: determining whether the first directed graph was updated; and in response to a determination that the first directed graph was update.
H-11. The medium of any of embodiments H-1 to H-10, wherein generating the integrated test condition comprises: determining a set of values associated with the shared resource type, wherein each of the set of values is used by at least one of the subset of conditional statements; and setting a test condition threshold as a maximum or minimum value of the set of values, wherein the integrated test condition comprises the test condition threshold.
H-12. The medium of any of embodiments H-1 to H-11, wherein the integrated test condition is a first integrated test condition, and wherein the subset of conditional statements is a second subset of conditional statements, the operations further comprising: obtaining a third set of conditional statements, wherein each conditional statement of the third set of conditional statements is associated with a vertex of a third directed graph; determining a second subset of conditional statements from the third set of conditional statements based on a second category selected from the set of mutually exclusive categories; generating a second integrated test condition based on the second subset of conditional statements, wherein the second integrated test condition; and determining whether the simulated event triggers the second integrated test condition.
H-13. The medium of any of embodiments H-1 to H-12, wherein determining the simulated event comprises determining the simulated event based on an entity identifier associated with the first entity.
H-14. The medium of any of embodiments H-1 to H-13, wherein: the first entity is associated with a first entity role; determining the simulated event comprises determining the simulated event based on an entity identifier associated with a third entity; and the third entity is not associated with the first entity role.
H-15. The medium of any of embodiments H-1 to H-14, wherein the integrated test condition comprises a time threshold, and wherein determining whether the simulated event satisfies the integrated test condition comprises determining whether a time value of the simulated event satisfies the time threshold.
H-16. The medium of any of embodiments H-1 to H-15, the operations further comprising determining whether a simulated outcome state caused by the simulated event indicates an activation of a second conditional statement, wherein the second conditional statement is triggered by a subsequent event, and wherein the subsequent event satisfies a third conditional statement of the first directed graph or the second directed graph.
H-17. The medium of any of embodiments H-1 to H-16, the operations further comprising: determining whether the first set of conditional statements has been updated; updating the subset of conditional statements based on the updated first set of conditional statements in response to a determination that the first set of conditional statements has been updated; updating the integrated test condition based on the updated subset of conditional statements; and determining whether the simulated event triggers the updated integrated test condition.
H-18. The medium of any of embodiments H-1 to H-17, the operations further comprising steps for determining the simulated event.
H-19. The medium of any of embodiments H-1 to H-18, the operations further comprising steps for determining whether the simulated event triggers the integrated test condition.
H-20. A method to perform the operations of any of the embodiments H-1 to H-19.
H-21. A system, comprising: one or more processors; and memory storing instructions that when executed by the processors cause the processors to effectuate operations comprising: the operations of any one of embodiments H-1 to H-19.
I-1. A tangible, non-transitory, machine-readable medium storing instructions that, when executed by a computing system, effectuate operations comprising: determining, with a computer system, a set of features associated in memory of the computer system with a set of vertices of a first directed graph, wherein a feature of the set of features is associated in memory of the computer system with a category type comprising a set of mutually exclusive categories; obtaining, with the computer system, a set of feature values associated with the set of vertices, wherein each respective vertex of set of vertices is associated with a respective subset of feature values, wherein: each feature value is associated with a feature of the set of features, and the respective subset of feature values comprise a respective category of the set of mutually exclusive categories; selecting, with the computer system, a first subset of features based on the set of feature values, wherein the selecting comprises: determining a plurality of candidate subsets of features; determining, with the computer system, a plurality of feature subset scores associated with the plurality of candidate subsets of features based on a category label selected from the set of mutually exclusive categories and the set of feature values; and selecting the first subset of features based on the plurality of feature subset scores; performing, with the computer system, a first operation to determine a set of extracted feature values, the first operation comprising: determining a set of input values by increasing a set of feature values associated with the first subset of features with a set of weights; and determining, with the computer system, the set of extracted feature values based on the set of input values, wherein the set of extracted feature values comprises a first multidimensional vector associated with the first directed graph; determining, with the computer system, a metric based on a distance between the first multidimensional vector and a second multidimensional vector of a second directed graph; determining, with the computer system, whether the metric satisfies a first threshold; and storing, with the computer system, the metric in persistent storage.
I-2. The medium of embodiment I-1, wherein determining the plurality of feature subset scores associated with the plurality of candidate subsets comprises: determining a first candidate subset of features, wherein the plurality of candidate subsets comprises the first candidate subset of features; determining a first feature subset score based on the first candidate subset of features using a neural network or decision tree; and selecting the first candidate subset of features as the first subset of features based on the first feature subset score being a maximum or minimum of the plurality of feature subset scores.
I-3. The medium of any of embodiments I-1 to I-2, the operations further comprising: obtaining a set of eigenvectors; and computing the first multidimensional vector based on a first set of feature values for a first vertex and the set of eigenvectors, wherein a sum of the set of eigenvectors when weighted by the first multidimensional vector satisfies a second threshold associated with the first set of feature values.
I-4. The medium of any of embodiments I-1 to I-3, wherein determining an extracted feature score comprises using a neural network that comprises a set of input layers and a set of output layers, wherein a count of the set of input layers is equal to a count of the set of output layers.
I-5. The medium of any of embodiments I-1 to I-4, wherein determining the metric comprises determining a Minkowski distance between the first multidimensional vector and a second multidimensional vector.
I-6. The medium of any of embodiments I-1 to I-5, the operations further comprising determining a first subset of vertices of the set of vertices based on the first subset of features and the set of extracted feature values satisfying a third threshold.
I-7. The medium of embodiment I-6, wherein determining the first subset of vertices comprises obtaining a set of prioritization parameters comprising the third threshold via a user interface element.
I-8. The medium of any of embodiments I-1 to I-7, the operations further comprising: determining whether a graph portion of the first directed graph matches a graph portion template, the graph portion template indicating a first vertex template, a second vertex template, and a directed edge template; generate an indicator associated with the graph portion; and visually indicating the graph portion associated with the graph portion based on the indicator.
I-9. The medium of any of embodiments I-1 to I-8, the operations further comprising increasing a feature value associated with a first feature in response to a determination that a first vertex is associated with a first category label and that a first conditional statement associated with the first vertex is satisfied.
I-10. The medium of any of embodiments I-1 to I-9, the operations further comprising visually indicating a natural language text section associated with a vertex of the first directed graph.
I-11. The medium of embodiment I-10, wherein determining the set of feature values comprises: determining a set of embedding values based on the natural language text section using a neural network; and determining a topic score based on the set of embedding scores, wherein the set of feature values comprises the topic score.
I-12. The medium of any of embodiments I-1 to I-11, the operations further comprising visually indicating a first vertex of the first directed graph with a shape, color, pattern, or animation that is different from a shape, color, pattern, or animation of a second vertex of the first directed graph in a visual display of the first directed graph.
I-13. The medium of any of embodiments I-1 to I-12, the operations further comprising providing a user interface, the user interface comprising a set of shapes representing vertices and a set of lines connecting the set of shapes, wherein each line is associated with an edge.
I-14. The medium of embodiment 1-13, the operations further comprising providing a user interface (UI), the UI indicating a first subset of vertices in a different color than a second subset of vertices of the first directed graph.
I-15. The medium of any of embodiments I-1 to I-14, wherein determining the set of extracted feature values comprises: determining a set of adjacency values associated with a first vertex; and determining a matrix multiplication product based on the set of adjacency values and one or more feature values of the set of feature values.
I-16. The medium of any of embodiments I-1 to I-15, the operations further comprising providing a user interface (UI), the UI comprising: a set of identifiers associated with the first subset of features; and a set of UI elements that, after manipulation, causes an update to a feature value of a vertex of the first directed graph.
I-17. The medium of embodiment 1-16, the operations further comprising determining a limit associated with a first feature of the first subset of features, wherein the UI causes a display of the limit.
I-18. The medium of any of embodiments I-1 to I-17, the operations further comprising steps for determining the metric between the first directed graph and the second directed graph.
I-19. The medium of any of embodiments I-1 to I-18, the operations further comprising steps for determining the first subset of features.
I-20. A method to perform the operations of any of the embodiments I-1 to I-19.
I-21. A system comprising: one or more processors; and memory storing instructions that, when executed by at least one of the one or more processors, causes at least one of the one or more processors to effectuate any of the operations of embodiments I-1 to I-19.
J-1. A tangible, non-transitory, machine-readable medium storing instructions that, when executed by a computer system, effectuate operations comprising: obtaining, with a computer system, program state of a self-executing protocol, wherein the program state comprises: a set of conditional statements; a first identifier of a first entity; and a directed graph, the directed graph comprising a set of vertices and a set of directed edges connecting respective pairs of vertices among the set of vertices, wherein each respective vertex of the set of vertices is associated with a respective category label of a set of mutually exclusive categories; receiving, at an application program interface of the computer system, an event message comprising a set of parameters; selecting, with the computer system, a first subset of vertices triggered by the event message based on the set of parameters; selecting, with the computer system, a second subset of vertices based on the first subset of vertices, wherein the second subset of vertices is associated with the first subset of vertices via the set of directed edges; determining, with the computer system, an aggregated parameter based on a subset of conditional statements, wherein each respective conditional statement of the subset of conditional statements is associated with a respective vertex of the second subset of vertices, and wherein the respective vertex is associated with a first category label of the set of mutually exclusive categories that is associated to each of the other vertices associated with the subset of conditional statements; and storing, with the computer system, the aggregated parameter in memory.
J-2. The medium of embodiment J-1, the operations further comprising: determining whether the event message is valid using a set of validator nodes of a peer-to-peer network, wherein each node of the peer-to-peer network is communicatively coupled to at least one other node of the peer-to-peer network; in response to a determination that the event message is valid, distributing a validation message indicating that the event message is valid; and storing a value based on the event message on a tamper-evident, distributed ledger encoding records of a plurality of previous values in a directed acyclic graph of cryptographic hash pointers, wherein the tamper-evident, distributed ledger is stored on the peer-to-peer network.
J-3. The medium of embodiment J-2, wherein determining the first subset of vertices comprises determining the first subset of vertices at a first node of the peer-to-peer network before the validation message is received by the first node.
J-4. The medium of any of embodiments J-1 to J-3, wherein the program state further comprises a first identifier of a first entity, the operations further comprising: determining whether the event message is valid using a set of validator nodes of a peer-to-peer network; based on a determination that the event message is not valid, sending an issue notification to a node of the peer-to-peer network associated with the first entity, wherein the issue notification comprises an identifier of the event message.
J-5. The medium of any of embodiments J-1 to J-4, wherein the program state further comprises a first identifier of a first entity, the operations further comprising: determining a network path from a first node of a peer-to-peer network to a second node of the peer-to-peer network using a breadth first search, wherein: the first node received the event message before the second node, the second node is associated with the first entity, and the network path comprises a plurality of nodes of the peer-to-peer network; and sending data of the event message to the second node from the first node via the network path.
J-6. The medium of any of embodiments J-1 to J-5, wherein the event message is a first event message, the operations further comprising: receiving a second event message within a duration threshold before or after receiving the first event message; determining whether the second event message causes a vertex of the first subset of vertices to trigger; in response to a determination that the second event message causes the vertex of the first subset of vertices to trigger, obtaining a set of triggering parameters of the second event message, wherein the set of triggering parameters comprise values that satisfy a condition of the vertex; determining whether a first value of the first event message and a second value of the second event message differ with respect to the set of triggering parameters; and based on a determination that the first value matches the second value, updating a parameter associated with the second event message to indicate that the second event message is a duplicate event message.
J-7. The medium of any of embodiments J-1 to J-6, wherein the program state further comprises a first identifier of a first entity and an second identifier of a second entity, the operations further comprising: retrieving a private conditional statement associated with the first entity, wherein the private conditional statement is not stored in program state accessible to the second entity; and determining whether the private conditional statement is satisfied based on the first subset of vertices or the second subset of vertices.
J-8. The medium of any of embodiments J-1 to J-7, wherein a first stored value of the self-executing protocol is stored on a peer-to-peer network, and wherein a first node of the peer-to-peer network is permitted to access the first stored value of the program state, and wherein a second node of the peer-to-peer network is not permitted to access the first stored value.
J-9. The medium of any of embodiments J-1 to J-8, wherein determining the aggregated parameter comprises: determining that triggering a first vertex of a pair of vertices of the directed graph causes the cancellation of a second vertex of the pair of vertices of the directed graph, wherein the first vertex is associated with a first conditional statement and the second vertex is associated with a second conditional statement; selecting one of the pair parameters, the pair of parameters comprising a first parameter of the first conditional statement and a second parameter of the second conditional statement; and determining the aggregated parameter based on the first parameter.
J-10. The medium of any of embodiments J-1 to J-9, wherein the program state further comprises a first identifier of a first entity, and wherein the first entity is associated with an entity role, the operations further comprising selecting the first entity, wherein selecting the first entity comprises: selecting a vertex of the first subset of vertices based on the set of parameters; and selecting the first entity based on the entity role being associated with the vertex.
J-11. The medium of embodiment J-10, wherein a second entity is associated the entity role, the operations further comprising sending a second message to the second entity based on the second entity being associated with the entity role.
J-12. The medium of any of embodiments J-1 to J-11, wherein the program state further comprises a first identifier of a first entity, the operations further comprising: determining that the first entity is associated with an entity role; in response to a determination that the first entity is associated with the entity role, selecting a previous message from a history of messages based on the entity role; and sending the previous message to the first entity.
J-13. The medium of any of embodiments J-1 to J-12, the operations further comprising providing a user interface (UI), wherein vertices displayed in the UI are colored based on color associations with category labels associated with the vertices, and wherein each respective category label of the set of mutually exclusive categories is associated with a different color.
J-14. The medium of any of embodiments J-1 to J-13, wherein the program state further comprises a first identifier of a first entity, the operations further comprising: determining whether a first confirmation key associated with a first representative of the first entity is received; determining whether a second confirmation key associated with a second representative the first entity is received; and in response to a determination that the first confirmation key and the second confirmation key is received, storing the first confirmation key and the second confirmation key in data storage in association with a record of a transaction between a pair entities comprising the first entity.
J-15. The medium of any of embodiments J-1 to J-14, wherein the program state further comprises a first identifier of a first entity, the operations further comprising; obtaining a score associated with the first entity, wherein the score is associated with a resource type; and updating the score based on the set of parameters, wherein the set of parameters comprises the resource type.
J-16. The medium of any of embodiments J-1 to J-15, wherein determining the aggregated parameter comprises determining a sum of values, wherein each respective value used to determine the sum of values is encoded in a respective conditional statement of the subset of conditional statements.
J-17. The medium of any of embodiments J-1 to J-16, the operations further comprising providing a user interface (UI), wherein the UI visually indicates the second subset of vertices based on a difference in color, difference in size, or difference in animation between the second subset of vertices and other vertices of the set of vertices.
J-18. The medium of any of embodiments J-1 to J-17, wherein determining the first subset of vertices comprises steps for determining the first subset of vertices.
J-19. The medium of any of embodiments J-1 to J-18, wherein determining the aggregated parameter comprises steps for determining the aggregated parameter.
J-20. A method to perform the operations of any of the embodiments J-1 to J-19.
J-21. A system comprising: one or more processors; and memory storing instructions that, when executed by at least one of the one or more processors, causes at least one of the one or more processors to effectuate any of the operations of embodiments J-1 to J-19.
K-1. A tangible, non-transitory, machine-readable medium storing instructions that, when executed by a computer system, effectuate operations comprising: receiving, with a computer system, a request via an application program interface (API), wherein the request comprises a callback address; determining, with the computer system, a query based on a set of query parameters; determining, with the computer system, a target graph portion template based on the query; searching, with the computer system, a set of directed graphs to determine a set of graph portions based on the query, where each of the set of graph portions match the target graph portion template, and wherein each respective directed graph of the set of directed graphs comprises: a set of vertices, wherein each respective vertex of the set of vertices is associated with a respective category label of a set of mutually exclusive categories, and a set of directed edges connecting respective pairs of vertices among the set of vertices; selecting, with the computer system, a set of event records, wherein each respective event records of the set of event records is indicated to occur before or during a vertex of a respective graph portion matching the target graph portion template; and sending, with the computer system, a value of the set of event records to the callback address.
K-2. The medium of embodiment K-1, wherein the query is associated with a first entity, the operations further comprising determining the set of related entities of the first entity, wherein each respective entity of the set of related entities is indicated to have had a transaction with the first entity.
K-3. The medium of any of embodiments K-1 to K-2, the operations further comprising determining whether the target graph portion template is stored in a library of graph portion templates, wherein: the library of graph portion templates comprises a graph database; a respective record of the graph database is associated with a respective graph portion template, and a respective identifier of the respective record comprises a respective set of vertices and a respective set of edges associating the respective set of vertices.
K-4. The medium of any of embodiments K-1 to K-3, the operations further comprising: determining whether a candidate graph portion matches with a graph portion template; updating a count associated with the graph portion template in a database based a determination that the candidate graph portion satisfies the graph portion template.
K-5. The medium of any of embodiments K-1 to K-4, the operations further comprising: determining whether the target graph portion template matches with a graph portion template stored in a library of graph portion templates; and in response to a determination that the target graph portion template does not matches the graph portion template stored in the library of graph portion templates, update the library of graph portion templates based on the target graph portion template.
K-6. The medium of any of embodiments K-1 to K-5, wherein the query is a first query, the operations further comprising: determining whether results of a set of previous searches based on a preceding query have been made, where the first query occurs after the preceding query; determining a first search time based on the preceding query, wherein the first search time indicates a time of occurrence for the set of previous searches; and modifying the query based on the first search time.
K-7. The medium of any of embodiments K-1 to K-6, wherein the set of vertices are encoded as a serialized array of vertices, and wherein determining a set of graph portions comprises: deserializing the serialized array of vertices to generate a first directed graph in a non-persistent memory, wherein the first directed graph encodes the set of vertices, set of entities, and set of directed edges; determining a first graph portion based on the first directed graph, wherein selecting the set of graph portions comprises selecting the first graph portion.
K-8. The medium of any of embodiments K-1 to K-7, wherein: a directed graph of the set of directed graphs is stored on a tamper-evident, distributed ledger encoding records of a plurality of previous values in a directed acyclic graph of cryptographic hash pointers, wherein the tamper-evident, distributed ledger is stored on a peer-to-peer network; each record comprises a relational database record, the relational database record comprising a balanced search tree (b-tree); and a set of root values of the b-tree comprise identifiers associated with vertices of the directed graph or graph portions of the directed graph.
K-9. The medium of embodiment K-8, the operations further comprising determining whether the query comprises at least one of a first set of query parameters, wherein searching through the set of directed graphs comprises using the b-tree in response to a determination that the query comprises at least one of the first set of query parameters.
K-10. The medium of any of embodiments K-1 to K-9, the operations further comprising: wherein each respective directed graph of the set of directed graphs is stored on a respective tamper-evident, distributed ledger encoding records of a plurality of previous values in a directed acyclic graph of cryptographic hash pointers, wherein the tamper-evident, distributed ledger is stored on a peer-to-peer network; storing a first version of a first directed graph of the set of directed graphs on a set of data centers, wherein the set of data centers does not use at least one peer node of the peer-to-peer network; and wherein searching through the set of directed graphs comprises searching through the first version of the first directed graph stored on the set of data centers.
K-11. The medium of embodiment K-10, wherein the set of data centers is a first set of data centers, the operation further comprising: storing a second version of the first directed graph on a second set of data centers, wherein the second set of data centers is different from the first set of data centers; and determining whether the first version of the first directed graph is valid based on the second version of the first directed graph.
K-12. The medium of any of embodiments K-1 to K-11, wherein determining the query comprises: determining whether an entity satisfies an access criteria, wherein the request is received from the entity; and based on a determination that the entity satisfies the access criteria, update the entity.
K-13. The medium of any of embodiments K-1 to K-12, wherein determining the query comprises: determining whether the set of query parameters comprises a command; and incorporating the command into the query.
K-14. The medium of any of embodiments K-1 to K-13, wherein the request is a first request, the operations further comprising: receiving a second request; determining whether the second request is valid; and in response to a determination that the second request is not valid, sending a message indicating that the second request is invalid.
K-15. The medium of any of embodiments K-1 to K-14, the operations further comprising: wherein each respective directed graph of the set of directed graphs is stored on a respective tamper-evident, distributed ledger encoding records of a plurality of previous values in a directed acyclic graph of cryptographic hash pointers, wherein the tamper-evident, distributed ledger is stored on a peer-to-peer network; receiving the request at a first node of the peer-to-peer network; determining whether the request causes a database search; in response to a determination that the request causes a database search, sending the request to a second node; wherein searching through the set of directed graphs comprises performing the database search using the second node.
K-16. The medium of any of embodiments K-1 to K-15, wherein searching through the set of directed graphs comprises searching through a set of binary trees associated with the set of directed graphs.
K-17. The medium of any of embodiments K-1 to K-16, the operations further comprising determining whether the callback address is valid based on a set of permitted addresses.
K-18. The medium of any of embodiments K-1 to K-17, wherein searching through the set of directed graphs comprises steps for searching through the set of directed graphs.
K-19. The medium of any of embodiments K-1 to K-18, wherein determining the query comprises steps for determining the query.
K-20. A method to perform the operations of any of the embodiments K-1 to K-19.
K-21. A system comprising: one or more processors; and memory storing instructions that, when executed by at least one of the one or more processors, causes at least one of the one or more processors to effectuate any of the operations of embodiments K-1 to K-19.
L-1. A tangible, non-transitory, machine-readable medium storing instructions that, when executed by a computer system, effectuate operations comprising: obtaining, with a computer system, a directed graph encoding a symbolic artificial intelligence (AI) model used by a first entity, the directed graph comprising a first set of vertices and a first set of edges associating pairs of vertices of the first set of vertices, wherein: each respective vertex of the first set of vertices is associated with a vertex status and is labeled with a category selected from a set of mutually-exclusive categories, and a vertex of the first set of vertices is associated with a conditional statement that is indicated as triggerable by the first entity; determining, with the computer system, a set of features based on the directed graph, the set of features comprising an identifier of a graph portion template, wherein each respective vertex of the graph portion template of the graph portion template is labeled with a same category from the set of mutually-exclusive categories as a corresponding respective vertex of a graph portion of the directed graph and is associated with a same count of edges; obtaining, with the computer system, a set of model parameter values for a machine learning model based on the graph portion template; and providing, with the computer system, the set of model parameter values and the graph portion templates to the first entity, wherein the set of model parameter values are used by to determine an outcome score based on the directed graph.
L-2. The medium of embodiment L-1, the operations further comprising: obtaining a conditional statement parameter, wherein the conditional statement parameter is used by the conditional statement; wherein obtaining the set of model parameter values comprises selecting a parameter of the set of model parameter values based on the conditional statement parameter.
L-3. The medium of any of embodiments L-1 to L-2, wherein the outcome score is a first outcome score, the operations further comprising: determining a map indicating a first graph portion of the directed graph, wherein the map comprises identifiers for vertices of the directed graph; and determining a second outcome score using the set of model parameters based on the map.
L-4. The medium of any of embodiments L-1 to L-3, wherein the directed graph is a first directed graph, and wherein determining the first graph portion comprises: using a first neural network based on the first directed graph and the set of model parameter values to determine the first outcome score, wherein the machine learning model comprises the first neural network; updating the map multiple times to generate a plurality of simulated directed graphs, wherein a first simulated directed graph matches the first graph portion with respect to vertex categories, and wherein each of the simulated directed graphs is a subgraph of the first directed graph; determining a set of simulated outcome scores with the neural network based on the plurality of simulated directed graphs and the set of model parameter values, wherein the plurality of simulated outcome scores comprises the second outcome score, and wherein using the neural network based on the first simulated directed graph provides the second outcome score; selecting the second outcome score of the set of simulated outcome scores based a difference between the second outcome score and the first outcome score; selecting the first graph portion based on the selection of the second outcome score; and sending an indicator associated with the first graph portion to the first entity.
L-5. The medium of any of embodiments L-1 to L-4, wherein the directed graph is a first directed graph, and wherein the set of features is a first set of features, and wherein determining the first graph portion comprises: using a first neural network based on the first directed graph and the set of model parameter values to determine the first outcome score, wherein the machine learning model comprises the first neural network; updating the map multiple times to generate a plurality of sets of features, wherein each respective set of features is different from the first set of features; determining a set of simulated outcome scores based on the plurality of sets of features using the set of model parameter values and the neural network, wherein the plurality of simulated outcome scores comprises the second outcome score, and wherein using the neural network based on a first set of features provides the second outcome score; selecting the second outcome score of the set of simulated outcome scores based a difference between the second outcome score and the first outcome score; selecting the first graph portion based on the first graph portion being associated with first set of features; and sending an indicator associated with the first graph portion to the first entity.
L-6. The medium of any of embodiments L-1 to L-5, the operations further comprising: obtaining a plurality of directed graphs; determining multiple sets of features, wherein each respective set of features is determined based on a respective directed graph of the plurality of directed graphs; determining the set of model parameter values by training a version of the machine learning model based on the multiple sets of features and a set of training outputs; storing the set of model parameter values in a record of a database, wherein the record is associated with a set of labels, and wherein a search through the database using the set of labels provides an identifier of the record.
L-7. The medium of any of embodiments L-1 to L-6, wherein determining the multiple sets of features comprises determining a set of values in a non-Euclidean domain based on the multiple sets of directed graphs, and wherein determining the set of values in the non-Euclidean domain comprises determining a matrix inversion of a diagonal of an adjacency matrix of the directed graph.
L-8. The medium of any of embodiments L-1 to L-7, the operations further comprising: obtaining a first transaction score, wherein the first transaction score is associated with a first transaction indicated to have triggered or activated a first vertex of a first directed graph of the plurality of directed graphs; obtaining a second transaction score, wherein the second transaction score is associated with a second transaction indicated to have triggered or activated a second vertex of a second directed graph of the plurality of directed graphs; aggregating the first score and the second score to form an aggregated score, the aggregating comprising a summation operation; and determining the set of model parameter values during a training operation based on the aggregated score.
L-9. The medium of any of embodiments L-1 to L-8, wherein obtaining the plurality of directed graphs comprises obtaining the plurality of directed graphs from a tamper-evident, distributed ledger.
L-10. The medium of any of embodiments L-1 to L-9, the operations further comprising selecting the machine learning model based on the set of features, wherein obtaining the set of model parameter values comprises selecting a model parameter value based on the selected learning model.
L-11. The medium of any of embodiments L-1 to L-10, wherein the set of features is a first set of features, the operations further comprising determining a second set of features based on the selected learning model and the directed graph, wherein obtaining the set of model parameter values comprises selecting a model parameter value based on the second set of features.
L-12. The medium of any of embodiments L-1 to L-11, wherein the first set of features and the second set of features are orthogonal to each other.
L-13. The medium of any of embodiments L-1 to L-12, wherein the set of model parameter values is received at a computer device controlled by the first entity, the operations further comprising: determining the outcome score using the set of model parameter values; and in response to a determination that the outcome score satisfies a warning threshold, sending a message to the first entity associated with the warning threshold.
L-14. The medium of any of embodiments L-1 to L-13, where obtaining the set of model parameter values comprises obtaining the set of model parameter values from a record stored on a cloud computing resource.
L-15. The medium of any of embodiments L-1 to L-14, wherein obtaining the set of model parameter values comprises: determining an entity role associated with the entity; and selecting a model parameter value of the set of model parameter values based on the entity role.
L-16. The medium of any of embodiments L-1 to L-15, wherein the machine learning model comprises a neural network, and wherein the model parameter values comprise weights for neurons of a neural network.
L-17. The medium of any of embodiments L-1 to L-16, wherein the set of model parameter values is a first set of model parameter values, the operations further comprising: obtaining a second set of model parameter values associated with an entity identifier; determining a second outcome score based on the second set of model parameter values; and based on a comparison between the first outcome score and the second outcome score, causing a transaction that updates a score associated with an entity identified by the entity identifier, wherein the score is stored in tamper-evident, distributed ledger.
L-18. The medium of any of embodiments L-1 to L-17, wherein determining the set of model parameter values comprises steps for determining the set of model parameter values.
L-19. The medium of any of embodiments L-1 to L-18, wherein determining the outcome score comprises steps for determining the outcome score.
L-20. A method to perform the operations of any of the embodiments L-1 to L-19.
L-21. A system, comprising: one or more processors; and memory storing instructions that when executed by the processors cause the processors to effectuate operations comprising: the operations of any one of embodiments L-1 to L-19.
L-22. A tangible, non-transitory, machine-readable medium storing instructions that, when executed by a computer system, effectuate operations comprising: obtaining, with a first computer system controlled by a first entity, a first training set, the first training set comprising a plurality of pairs of symbolic AI models and logs of state of the symbolic AI models; training, with the first computer system, a first machine learning model on the first training set, wherein training comprises iteratively adjusting parameters of the first machine learning model based on a first objective function; after training the first machine learning model, providing the first machine learning model to a second computer system controlled by a second entity, wherein the second entity does not have access to at least some of the first training set; obtaining, with the second computer system, a second training set, the second training set comprising a plurality of pairs of symbolic AI models and logs of state of the symbolic AI models, the second training set being different at least in part from the first training set; training, with the second computer system, a second machine learning model that includes the first machine learning model on the second training set, wherein training comprises iteratively adjusting parameters of the second machine learning model based on a second objective function; and after training the second machine learning model, storing, with the second computer system, the second machine learning model in memory.
L-23. The medium of embodiment L-22, wherein the second machine learning model comprises: the first machine learning model as a sub-model having an output; an error-correcting machine learning model as a sub-model having an input based on the output of the first machine learning model, wherein training the second machine learning model comprises iteratively adjusting parameters of the error-correcting machine learning model without changing parameters of the of the first machine learning model while adjusting parameters of the error-correcting machine learning model.
L-24. The medium of any of embodiments L-22 to L-23, wherein: parameters of the second machine learning model are initialized to values of parameters of the first machine learning model before being iteratively adjusted during training of the second machine learning model.
L-25. The medium of any of embodiments L-22 to L-24, wherein the first and second machine learning models are non-symbolic artificial intelligence models comprising steps for machine learning.
L-26. A method to perform the operations of any of the embodiments L-22 to L-25.
L-27. A system, comprising: one or more processors; and memory storing instructions that when executed by the processors cause the processors to effectuate operations comprising: the operations of any one of embodiments L-22 to L-26.
Claims
1. A tangible, non-transitory, machine-readable medium storing instructions that, when executed by a computer system, effectuate operations comprising:
- obtaining, with a computer system, a directed graph encoding a symbolic artificial intelligence (AI) model used by a first entity, the directed graph comprising a first set of vertices and a first set of edges associating pairs of vertices of the first set of vertices, wherein: each respective vertex of the first set of vertices is associated with a vertex status and is labeled with a category selected from a set of mutually-exclusive categories, and a vertex of the first set of vertices is associated with a conditional statement that is indicated as triggerable by the first entity;
- determining, with the computer system, a set of features based on the directed graph, the set of features comprising an identifier of a graph portion template, wherein each respective vertex of the graph portion template of the graph portion template is labeled with a same category from the set of mutually-exclusive categories as a corresponding respective vertex of a graph portion of the directed graph and is associated with a same count of edges;
- obtaining, with the computer system, a set of model parameter values for a machine learning model based on the graph portion template; and
- providing, with the computer system, the set of model parameter values and the graph portion templates to the first entity, wherein the set of model parameter values are used by to determine an outcome score based on the directed graph.
2. The medium of claim 1, the operations further comprising:
- obtaining a conditional statement parameter, wherein the conditional statement parameter is used by the conditional statement;
- wherein obtaining the set of model parameter values comprises selecting a parameter of the set of model parameter values based on the conditional statement parameter.
3. The medium of claim 1, wherein the outcome score is a first outcome score, the operations further comprising:
- determining a map indicating a first graph portion of the directed graph, wherein the map comprises identifiers for vertices of the directed graph; and
- determining a second outcome score using the set of model parameters based on the map.
4. The medium of claim 3, wherein the directed graph is a first directed graph, and wherein determining the first graph portion comprises:
- using a first neural network based on the first directed graph and the set of model parameter values to determine the first outcome score, wherein the machine learning model comprises the first neural network;
- updating the map multiple times to generate a plurality of simulated directed graphs, wherein a first simulated directed graph matches the first graph portion with respect to vertex categories, and wherein each of the simulated directed graphs is a subgraph of the first directed graph;
- determining a set of simulated outcome scores with the neural network based on the plurality of simulated directed graphs and the set of model parameter values, wherein the plurality of simulated outcome scores comprises the second outcome score, and wherein using the neural network based on the first simulated directed graph provides the second outcome score;
- selecting the second outcome score of the set of simulated outcome scores based a difference between the second outcome score and the first outcome score;
- selecting the first graph portion based on the selection of the second outcome score; and
- sending an indicator associated with the first graph portion to the first entity.
5. The medium of claim 3, wherein the directed graph is a first directed graph, and wherein the set of features is a first set of features, and wherein determining the first graph portion comprises:
- using a first neural network based on the first directed graph and the set of model parameter values to determine the first outcome score, wherein the machine learning model comprises the first neural network;
- updating the map multiple times to generate a plurality of sets of features, wherein each respective set of features is different from the first set of features;
- determining a set of simulated outcome scores based on the plurality of sets of features using the set of model parameter values and the neural network, wherein the plurality of simulated outcome scores comprises the second outcome score, and wherein using the neural network based on a first set of features provides the second outcome score;
- selecting the second outcome score of the set of simulated outcome scores based a difference between the second outcome score and the first outcome score;
- selecting the first graph portion based on the first graph portion being associated with first set of features; and
- sending an indicator associated with the first graph portion to the first entity.
6. The medium of claim 1, the operations further comprising:
- obtaining a plurality of directed graphs;
- determining multiple sets of features, wherein each respective set of features is determined based on a respective directed graph of the plurality of directed graphs;
- determining the set of model parameter values by training a version of the machine learning model based on the multiple sets of features and a set of training outputs;
- storing the set of model parameter values in a record of a database, wherein the record is associated with a set of labels, and wherein a search through the database using the set of labels provides an identifier of the record.
7. The medium of claim 6, wherein determining the multiple sets of features comprises determining a set of values in a non-Euclidean domain based on the multiple sets of directed graphs, and wherein determining the set of values in the non-Euclidean domain comprises determining a matrix inversion of a diagonal of an adjacency matrix of the directed graph.
8. The medium of claim 6, the operations further comprising:
- obtaining a first transaction score, wherein the first transaction score is associated with a first transaction indicated to have triggered or activated a first vertex of a first directed graph of the plurality of directed graphs;
- obtaining a second transaction score, wherein the second transaction score is associated with a second transaction indicated to have triggered or activated a second vertex of a second directed graph of the plurality of directed graphs;
- aggregating the first score and the second score to form an aggregated score, the aggregating comprising a summation operation; and
- determining the set of model parameter values during a training operation based on the aggregated score.
9. The medium of claim 6, wherein obtaining the plurality of directed graphs comprises obtaining the plurality of directed graphs from a tamper-evident, distributed ledger.
10. The medium of claim 1, further comprising selecting the machine learning model based on the set of features, wherein obtaining the set of model parameter values comprises selecting a model parameter value based on the selected learning model.
11. The medium of claim 10, wherein the set of features is a first set of features, the operations further comprising determining a second set of features based on the selected learning model and the directed graph, wherein obtaining the set of model parameter values comprises selecting a model parameter value based on the second set of features.
12. The medium of claim 11, wherein the first set of features and the second set of features are orthogonal to each other.
13. The medium of claim 1, wherein the set of model parameter values is received at a computer device controlled by the first entity, the operations further comprising:
- determining the outcome score using the set of model parameter values; and
- in response to a determination that the outcome score satisfies a warning threshold, sending a message to the first entity associated with the warning threshold.
14. The medium of claim 1, where obtaining the set of model parameter values comprises obtaining the set of model parameter values from a record stored on a cloud computing resource.
15. The medium of claim 1, wherein obtaining the set of model parameter values comprises:
- determining an entity role associated with the entity; and
- selecting a model parameter value of the set of model parameter values based on the entity role.
16. The medium of claim 1, wherein the machine learning model comprises a neural network, and wherein the model parameter values comprise weights for neurons of a neural network.
17. The medium of claim 1, wherein the set of model parameter values is a first set of model parameter values, the operations further comprising:
- obtaining a second set of model parameter values associated with an entity identifier;
- determining a second outcome score based on the second set of model parameter values; and
- based on a comparison between the first outcome score and the second outcome score, causing a transaction that updates a score associated with an entity identified by the entity identifier, wherein the score is stored in tamper-evident, distributed ledger.
18. The medium of claim 1, wherein determining the set of model parameter values comprises steps for determining the set of model parameter values.
19. The medium of claim 1, wherein determining the outcome score comprises steps for determining the outcome score.
20. A method comprising:
- obtaining, with a computer system, a directed graph encoding a symbolic artificial intelligence (AI) model used by a first entity, the directed graph comprising a first set of vertices and a first set of edges associating pairs of vertices of the first set of vertices, wherein: each respective vertex of the first set of vertices is associated with a vertex status and is labeled with a category selected from a set of mutually-exclusive categories, and a vertex of the first set of vertices is associated with a conditional statement that is indicated as triggerable by the first entity;
- determining, with the computer system, a set of features based on the directed graph, the set of features comprising an identifier of a graph portion template, wherein each respective vertex of the graph portion template of the graph portion template is labeled with a same category from the set of mutually-exclusive categories as a corresponding respective vertex of a graph portion of the directed graph and is associated with a same count of edges;
- obtaining, with the computer system, a set of model parameter values for a machine learning model based on the graph portion template; and
- providing, with the computer system, the set of model parameter values and the graph portion templates to the first entity, wherein the set of model parameter values are used by to determine an outcome score based on the directed graph.
21. A tangible, non-transitory, machine-readable medium storing instructions that, when executed by a computer system, effectuate operations comprising:
- obtaining, with a first computer system controlled by a first entity, a first training set, the first training set comprising a plurality of pairs of symbolic AI models and logs of state of the symbolic AI models;
- training, with the first computer system, a first machine learning model on the first training set, wherein training comprises iteratively adjusting parameters of the first machine learning model based on a first objective function;
- after training the first machine learning model, providing the first machine learning model to a second computer system controlled by a second entity, wherein the second entity does not have access to at least some of the first training set;
- obtaining, with the second computer system, a second training set, the second training set comprising a plurality of pairs of symbolic AI models and logs of state of the symbolic AI models, the second training set being different at least in part from the first training set;
- training, with the second computer system, a second machine learning model that includes the first machine learning model on the second training set, wherein training comprises iteratively adjusting parameters of the second machine learning model based on a second objective function; and
- after training the second machine learning model, storing, with the second computer system, the second machine learning model in memory.
22. The medium of claim 21, wherein the second machine learning model comprises:
- the first machine learning model as a sub-model having an output;
- an error-correcting machine learning model as a sub-model having an input based on the output of the first machine learning model, wherein training the second machine learning model comprises iteratively adjusting parameters of the error-correcting machine learning model without changing parameters of the of the first machine learning model while adjusting parameters of the error-correcting machine learning model.
23. The medium of claim 21, wherein:
- parameters of the second machine learning model are initialized to values of parameters of the first machine learning model before being iteratively adjusted during training of the second machine learning model.
24. The medium of claim 21, wherein the first and second machine learning models are non-symbolic artificial intelligence models comprising steps for machine learning.
Type: Application
Filed: Sep 8, 2020
Publication Date: Mar 11, 2021
Inventor: Edward Hunter (Gaithersburg, MD)
Application Number: 17/015,074
