Abstract: A model deployment system (MDS) is disclosed that includes capabilities for generating an optimized model that is tailored to specific target deployment hardware used to deploy the model in a target environment. The MDS is capable of executing a model inference accelerator (MIA) layer to perform a set of optimizations on a machine learning (ML) model. The MIA layer executes the set of optimizations based on model metadata associated with the ML model to generate an optimized ML model. The execution performed by the MIA layer includes generating an intermediate representation (IR) format for the IR model that identifies model-specific optimizations to be applied to the ML model and target hardware for deploying the ML model. The execution by the MIA layer further includes compiling the IR format of the ML model based on model-specific optimizations and target hardware-specific model optimizations to generate an optimized ML model.

Abstract: Techniques for detecting inter-tenancy exfiltration of data in a cloud environment are disclosed. A log that includes information associated with receipt of a service message at a gateway within a cloud environment is accessed. Based on the log, (i) originating information of the service message (such as identification of an originating tenancy of the service message) and (ii) target information of the service message (such as identification of a target tenancy of the service message) are determined. The originating information and the target information are compared. A mismatch between the originating information of the service message and the target information of the service message is detected. In response to the detected mismatch between the originating information of the service message and the target information of the service message, information indicative of the detected mismatch is caused to be presented at a user interface.

Abstract: Techniques for detecting anomalous domain name system (DNS) query semantics are disclosed. A DNS query, which is to be routed to a DNS server, is received. The DNS query is parsed. Sensitive information within the DNS query is detected. In response to detecting the sensitive information within the DNS query, the DNS query is flagged as including the sensitive information. In an example, in response to flagging the DNS query as including the sensitive information, the DNS query is blocked from being transmitted to its target destination.

Abstract: Techniques are disclosed for enabling dynamic routing of identity and access (IA) requests to a cloud identity and access management (IAM) system. In some embodiments, the health (or service quality) of a cloud IAM system is continuously monitored in the background to update routing information for dynamically routing IA requests for client cloud services to the cloud IAM system in the foreground. In some embodiments, background monitoring is performed by sending first-type IA requests to the cloud IAM system to discover traffic patterns. In some embodiments, routing information may be mappings between client cloud services and identity endpoints of service cells in the cloud IAM system with configurable attributes that help rebalance and distribute the IA traffic.

Abstract: An augmentation and feature caching subsystem is described for training AI/ML models. In one particular aspect, a method is provided that includes receiving data comprising training examples, one or more augmentation configuration hyperparameters and one or more feature extraction configuration hyperparameters; generating a first key based on one of the training examples and the one or more augmentation configuration hyperparameters; searching a first key-value storage based on the first key; obtaining one or more augmentations based on the search of the first key-value storage; applying the obtained one or more augmentations to the training examples to result in augmented training examples; generating a second key based on one of the augmented training examples and the one or more feature extraction configuration hyperparameters; searching a second key-value storage based on the second key; obtaining one or more features based on the search of the second key-value storage.

Abstract: A cloud service is provided and accessible via a virtual IP address in a virtual cloud network (VCN), where the VCN is part of an overlay network. A source network virtualization device (NVD) receives a first packet destined to the virtual IP address. The source NVD determined based upon a dynamic routing gateway table, the VCN mapped to the virtual IP address. Further, the source NVD identifies based upon the virtual IP address, a substrate IP address of a destination NVD that implements a virtual network interface card (VNIC) associated with the virtual IP address and communicates the first packet from the source NVD to the destination NVD using the substrate IP address.

Abstract: Disclosure relates to training a model to score video clips based on input prompts and associated ground truth information. Training datapoints may include a training video clip, an input prompt, and ground truth information. The model may generate a score for each clip based at least in part on the input prompt. A loss calculation subsystem may compute a loss using the score and the ground truth information. A loss minimization subsystem may update model parameters to train the model. In some embodiments, similarity scores may be generated for frames or images of a training video clip and/or be adjusted using a Weibull-based methodology to produce an overall clip score. In some embodiments, training may use an N-pair contrastive loss such that a clip annotated with a central time may receive a highest score while boundary clips annotated with a start time and/or end time retain partial similarity.

Abstract: Techniques for hybrid retrieval using a knowledge base (KB) system for responsive multi-turn conversations are disclosed. The system evaluates user queries by generating confidence scores from comparisons against a set of predefined intents. When a confidence score falls within a high-confidence range, the system produces a predefined intent-based response. When a confidence score falls within a low-confidence range, the system retrieves a relevant KB document and generates a response by applying a language model (LM) to the query and the retrieved document.

Abstract: Techniques are disclosed herein for generating augmented training examples to fine-tune a machine learning model for converting natural language to logical form. In one aspect, a method is disclosed that includes: accessing a set of training examples, wherein each training example includes: (i) an original prompt having a database schema and (ii) a programming language completion corresponding to the original prompt, generating a set of augmented training examples based on the set of training examples, generating a fine-tuning dataset comprising the training examples and the augmented training examples, and fine-tuning a model with the fine-tuning dataset.

Abstract: Techniques for configuring a machine learning (ML) pipeline are disclosed. A system generates a region comprehension index (RCI) value that quantifies, for a visual dataset supplied to a computer vision model (CVM), a bias toward global or local reasoning. The system determines a first performance metric at least by applying the CVM to a whole image included in the visual dataset and a second performance metric at least by applying the CVM to patches of the whole image. The system determines the RCI value as a function of at least the first performance metric and the second performance metric. If the RCI value satisfies a criterion for deploying the CVM to an ML pipeline, the system selectively deploys the CVM to the ML pipeline.

Abstract: Techniques for data communication protocols for data transmission and software assurance in a cloud environment are disclosed. Data is received from a first component and at a first adapter using a first communication protocol. The data is transmitted, by the first adapter and to a proxy, using a second communication protocol. The proxy audits the data, and (i) blocks passage of the data, or (ii) forwards the data to a second adapter using the second communication protocol, to cause the second adapter to transmit the data to a second component using the first communication protocol. The proxy is within a tenancy of a cloud environment. The first adapter and the first component are within a first section of the cloud environment. The second adapter and the second component are within a second section of the cloud environment. Neither of the first or second sections is within the tenancy.

Abstract: Techniques are provided for a clustering algorithm to harvest in-context examples from Text-to-SQL logs. The method includes receiving vector representations of natural language utterances and corresponding queries in a programming query language. The clustering algorithm may be reparametrized using a training dataset by sampling the vector representations, estimating local radius parameter values from a distribution of pairwise distances, and iteratively executing the clustering algorithm while incrementally relaxing density constraints to identify paraphrases and unique queries. Hyperparameter tuning may be performed by adjusting clustering parameters based on evaluation metrics to obtain a fine-tuned algorithm. The method can further augment examples with ground truth examples and execute the fine-tuned clustering algorithm on the augmented set to generate clustered data points representing groups of paraphrases and unique queries.

Abstract: Techniques for conducting membership inference attacks are disclosed. In an example, a plurality of target interactions of a target user with an item providing platform are monitored. A plurality of target recommendations for the target user is received from a recommendation system of the item providing platform. Using an attack classifier and based on (i) the plurality of target interactions and (ii) the plurality of target recommendations, an inference is made as to whether at least a subset of the plurality of target interactions and/or at least a subset of the plurality of target recommendations were used to train the recommendation system. The attack classifier is trained using training data associated with a plurality of autonomous users (such as autonomous sock puppets) interacting with the item providing platform. In an example, the item providing platform is one of a video providing platform, an audio providing platform, or a shopping platform.

Abstract: Techniques for undertaking containment actions in response to detection of stealing of non-user principals within a cloud environment are disclosed. Usage of a non-user principal by a first non-user entity is detected within a cloud environment, wherein the non-user principal was originally assigned to a second non-user entity that is different from the first non-user entity. A detection is made as to whether the non-user principal is assigned to a dynamic group of non-user entities within a tenancy. A detection is made as to whether the non-user principal is an instance principal, a resource principal, or a service principal. One or more containment actions are undertaken, based at least in part on one or both of (i) detecting whether the non-user principal is assigned to the dynamic group of non-user entities and (ii) detecting whether the non-user principal is an instance principal, a resource principal, or a service principal.

Abstract: Techniques are disclosed herein for generating a distillation dataset to train a machine learning model for converting natural language to logical form. In one aspect, a method is disclosed that includes: accessing customization information comprising a set of reference natural language queries, generating a set of synthetic natural language queries based on the customization information and database schema information and generating a set of distillation training examples. Generating the set of distillation training examples can include generating a synthetic programming language query for each synthetic natural language query. Each distillation training example comprises (i) a synthetic natural language query, (ii) a database schema from the database schema information associated with the synthetic natural language query, and (iii) a synthetic programming language query corresponding with the synthetic natural language query.

Abstract: A computer-implemented method for evaluating agenda-driven conversational agents is disclosed. The method includes receiving a natural language utterance associated with a predefined agenda and determining, by an agent generative model, whether clarification is required. If clarification is needed, a simulated conversation is executed between the agent generative model and a mimic generative model, where each turn involves generating, by the agent generative model, one or more clarifying questions and generating, by the mimic generative model, one or more clarifying responses based on the agenda. The agent generative model generates a response to the natural language utterance based on the conversation. The agent generative model's performance in generating clarifying questions and responses is evaluated, enabling objective assessment of the agent's ability to handle complex, multi-turn interactions using agenda-driven simulated dialogue.

Abstract: The present disclosure relates to centralized resource identification techniques, leveraging a standardized resource directory for storing resource information in a cloud network. The centralized resource identification techniques may be provided for creating a data ingestion layer to store and map data that may include mapping a resource ID to a server endpoint in a meta database, parsing the data to detect duplicate resources, validating the data, defining multiple APIs for the resources, registering errors, and updating the APIs. The techniques may reroute the requests from one server endpoint to another by receiving the request from a user, authenticating the request at an API gateway, retrieving an API of a resource associated with the request, and rerouting the request based on associated mapping and policies. The request processing may be automated using a machine learning engine, and one or more resources and their services may be displayed on a user interface.

Abstract: A computer obtains a trained conditional random field (CRF) model including a set of model parameters learned from training data and stored in a transition matrix. Tag sequences, inconsistent with the tag sequence logic, are identified for the tags within the transition matrix, setting, within the transition matrix, a cost associated with transitioning between the pair of tags to be equal to a predefined hyperparameter value that penalizes the transitioning between the inconsistent pair of tags. The CRF model receives a string of text including one or more named entities. The CRF model inputs the string of text into the CRF model having the cost associated with the transitioning between the pair of tags set equal to the predefined hyperparameter value. The CRF model classifies the words within the string of text into different classes which might include the one or more named entities.

Abstract: Example systems, methods, and computer-program products are described for using multiple AI agents to simulate performer-evaluator interaction and generate constructive feedback to automate pipeline progression. An example method includes a performer agent accessing performer-agent-accessible data potentially relevant to a project. The example method also includes the performer agent generating a performer recommendation to explain, from a performer perspective and with reference to at least some of the performer-agent-accessible data, why the project should advance in status according at least to stored objectives. The example method also includes an evaluator agent generating an evaluator recommendation to explain, from an evaluator perspective and with reference to at least some of the performer-agent-accessible data, why the project should not advance in status according at least to the stored objectives.

Abstract: The disclosed systems, methods, and computer readable media relate to sidelining a data stream partition (e.g., halting replication of the partition) and resuming replication for a sidelined partition after an a poison pill event that triggered the sidelining is resolved. While the partition is sidelined, stream event processing for the partition may be restricted/blocked/halted, while stream event processing for active partitions of the data stream continue. After the error is resolved, a process for resynchronizing the partition based on the stream events that were received from the point at which the error was originally detected may be performed. Some stream events for the partition may be obtained from a snapshot generated by a snapshot service, while other stream events during the resynchronization may be obtained from a demuxer. Once the partition is resynchronized, the partition may be un-sidelined such that subsequent stream events for the partition may be processed unrestricted.

Abstract: Techniques for a multi-level authentication within a cloud environment are disclosed. A first authentication request is received by an identity and access management (IAM) service from a device associated with a user. The first request is to authenticate the user for accessing one or more cloud resources through a gateway of the cloud environment. Responsive to the first authentication request, a redirection is performed to submit a second authentication request to an identity provider (IdP) to authenticate the user. A first token indicating a first identity for the user based on the first authentical is received by the IAM service, here the first authentication was performed by the IdP. The IAM service performs a second authentication of the user. The IAM service issues a second token indicating a second identity for the user. The device gains access to the one or more cloud resources, based on the second token.

Abstract: Disclosed are approaches for automatically generating and refining instructions for converting natural language questions to a logical form such as structured query language (SQL). An initial evaluation using a baseline prompt identifies errors in generated SQL queries. These errors are clustered to identify common failure patterns. For a selected error cluster, candidate instructions are generated to address the pattern. The candidate instructions are then validated to confirm they improve accuracy for the error cluster without degrading overall performance across a training set. Upon successful validation, the instructions are incorporated. This process of identifying error clusters, generating instructions, and validating them is repeated iteratively to progressively enhance the instruction set. The resulting instructions may also be automatically compressed to reduce length while maintaining performance, thereby improving operational efficiency.

Abstract: Techniques for configuring a machine learning (ML) pipeline are disclosed. One or more embodiments generate a patch context robustness index (PCRI) value that quantifies, for computer vision model (CVM), sensitivity of the CVM to contextual noise, at least by: determining a first performance metric at least by applying the CVM to a whole image; determining a second performance metric at least by applying the CVM to patches of the whole image; and determining the PCRI value as a function of at least the first performance metric and the second performance metric. One or more embodiments deploy the CVM to an ML pipeline based at least on the PCRI value.

Abstract: Techniques for detected reuse of a public key of a public/private cryptographic key pair within a cloud environment are disclosed. A database maintained within a cloud environment is accessed, the database storing a plurality of public keys. Each public key has a corresponding attribute, and the database further stores a plurality of attributes corresponding to the plurality of public keys. Each public key is associated with either (i) a corresponding user account of a corresponding tenancy of the cloud environment or (ii) a compute instance hosted within a corresponding tenancy of the cloud environment. A number of times a first public key or a first attribute corresponding to the first public key occurring within the database is determined. Responsive to the number of times being equal to or greater than a threshold number, the first public key is tagged as being reused for at least the threshold number of times.

Abstract: A source network virtualization device (NVD) receives a data packet destined for a cloud service accessible via a virtual IP address in a virtual cloud network (VCN). The source NVD encapsulates the data packet with a first encapsulation header to generate a first encapsulated data packet, the first encapsulated data packet being forwarded to a dynamic routing gateway (DRG). The DRG encapsulates the first encapsulated data packet with a second encapsulation header to generate a second encapsulated data packet, the second encapsulated data packet being forwarded to destination NVD associated with the cloud service, whereas the destination NVD decapsulates the second encapsulated data packet to obtain the first encapsulated data packet. The destination NVD communicates the first encapsulated data packet to the cloud service, wherein a virtual network interface card (VNIC) associated with the cloud service decapsulates the first encapsulated data packet to obtain the data packet.

Abstract: Techniques are disclosed for deploying a computing resource (e.g., a service) in response to user input. A computer-implemented method can include operations of receiving (e.g., by a gateway computer of a cloud-computing environment) a request comprising an identifier for a computing component of the cloud-computing environment. The computing device receiving the request may determine whether the identifier exists in a routing table that is accessible to the computing device. If so, the request may be forwarded to the computing component. If not, the device may transmit an error code (e.g., to the user device that initiated the request) indicating the computing component is unavailable and a bootstrap request to a deployment orchestrator that is configured to deploy the requested computing component. Once deployed, the computing component may be added to a routing table such that subsequent requests can be properly routed to and processed by the computing component.

Abstract: Techniques are provided for automated migration replication. A method can include creating a volume group including an initial snapshot of a virtual machine (VM) residing in an initial environment and in an initial configuration (VM1). The method can include generating a terraform stack based on the initial snapshot, execution of which in an environment causes replication of VM1 in that environment. The method can include providing the terraform stack to a user, generating a subsequent snapshot of the VM in a subsequent configuration (VM2), generating a delta file characterizing the difference between the initial snapshot stored in the volume group and the subsequent snapshot, generating a delta terraform stack based on the delta file, wherein execution of the delta terraform stack causes previously replicated VM1 to update to replicate VM2, and providing the delta terraform stack to the user.

Abstract: Techniques for de-centralized execution of workflows are disclosed. A system executes a sequence of jobs in a workflow. Each job includes application code to perform a defined set of functions. Each job includes wrapper code at one or both ends of the application code. The system executes the sequence of jobs by iteratively spinning up a virtual machine, loading workflow data to be used by application code, executing the application code, and storing modified workflow data. The virtual machine executing one job triggers the next virtual machine to execute the next job in the workflow, without scheduling the next job by a centralized job scheduler. Upon spinning up the next virtual machine, a virtual machine executing a preceding job shuts itself down.

Abstract: A method for detecting a deserialization attack may include identifying, in a byte stream, a class name corresponding to a class, generating, for the class, a feature vector, generating, by applying a benign deserialization model to the feature vector, a benign probability window, generating, by applying a malicious deserialization model to the feature vector, a malicious probability window, comparing the benign probability window and the malicious probability window to obtain a comparison result, and determining, based on the comparison result, that the class is malicious.

Abstract: The present disclosure relates to machine learning techniques for In-Context-Learning (ICL) with pattern-based retrieval for the task of converting Natural Language (NL) to Structured Query Language (SQL). Aspects are directed towards acquiring a natural language utterance and a database schema, searching, using at least a portion of the natural language utterance as a key, a memory bank for one or more in-context examples that are relevant to the key, generating a prompt comprising the natural language query, the database schema, and the one or more in-context examples, transmitting the prompt to a first pretrained generative artificial intelligence model, receiving, from the first pretrained generative artificial intelligence model, a logical form corresponding to the natural language utterance based at least in part on the prompt, executing the logical form on a database to obtain a query result, and providing the query result to a user.

Abstract: Disclosed is a system, method, and computer program product for implementing a log analytics method and system that can configure, collect, and analyze log records in an efficient manner. Machine learning-based classification can be performed to classify logs. This approach is used to group logs automatically using a machine learning infrastructure.

Abstract: Embodiments described herein are generally related to data analytics environments, and are particularly directed to systems and methods for use with a data analytics environment to provide enrichment of data from external sources via a large language model. In accordance with an embodiment, the systems and methods can utilize a high-powered LLM to suggest data-related factors that may not be explicitly represented in the user's dataset. For example, upon a user's selection of a pair of datapoints, the systems and methods can utilize a LLM to provide suggestions for root causes of those datapoints, even though such root causes are not explicitly represented in the dataset.

Abstract: Systems and methods are disclosed to improve disaster recovery by implementing a scalable low-loss disaster recovery for a data store. The disaster recovery system enables disaster recovery for a linearizable (e.g., externally consistent) distributed data store. The disaster recovery system also provides for a small lag on the backup site relative to the primary site, thereby reducing the data loss by providing a smaller data loss window compared to traditional disaster recovery techniques. The disaster recovery system implements a timestamp for log records based on a globally synchronized clock. The disaster recovery system also implements a watermark service that updates a global watermark timestamp that a backup node uses to apply log records.

Abstract: A system manager is disclosed for automatically deploying software based on user activity. The system manager receives user-specified parameters to control whether, when, and/or how automatic deployment is performed for software components, without specifying which software components are to be automatically deployed for the user. The system manager stores software component metadata identifying software components that are available for deployment. The system manager analyzes user activity to determine autodeployment utility scores that predict degrees of utility for different software components. Based on determining that an autodeployment utility score for a candidate software component satisfies one or more conditions that are based on the user-specified parameters, the system manager automatically performs deployment operation(s) for the candidate software component in a computing environment of a user.

Abstract: A first set and a second set are identified as operands for a set operation of a similarity analysis task iteration. Using respective minimum hash information arrays and contributor count arrays of the two sets, a minimum hash information array and contributor count array of a derived set resulting from the set operation is generated. An entry in the contributor count array of the derived set indicates the number of child sets of the derived set that meet a criterion with respect to a corresponding entry in the minimum hash information array of the derived set. The generated minimum hash information array and the contributor count array are stored as part of input for a subsequent iteration. After a termination criterion of the task is met, output of the task is stored.

Abstract: A system synchronizes a server-side DOM tree and a browser-side DOM tree with one another. Server may receive from a browser, a hash value of the browser-side DOM tree, and a server-side update instruction for applying a first server-side update to the server-side DOM tree to synchronize with a first browser-side update by the browser to the browser-side DOM tree. The server may identify the server-side DOM tree based on the hash value. The server may execute upon the server-side DOM tree, the first server-side update and a second server-side update that is triggered by the first server-side update. The server may compute a browser-side update instruction for applying a second browser-side update to the browser-side DOM tree to synchronize with the server-side DOM tree. The server may transmit the browser-side update instruction to the browser, and the browser may apply the second browser-side update to the browser-side DOM tree.

Abstract: Techniques are disclosed herein for generating dynamic vocabularies for conditioning a language model. A dynamic vocabulary is constructed from an input prompt, database schema information for a database to be queried, and programming language information for a programming language to be used for querying the database to condition the language model to predict an output statement in the programming language. The dynamic vocabulary can be included in prompt information that is provided to the language model. The number of tokens in the dynamic vocabulary can be different than a number of tokens included in a vocabulary of the language model. By utilizing a dynamic vocabulary, the language model can be conditioned to predict tokens for the output statement that are contextually consistent with the tokens included the dynamic vocabulary.

Abstract: Techniques for managing remote devices based on polling requests are disclosed. A system manages diagnostics, updating, and/or repair of client device. A device manager identifies the diagnostics and validation tests and other operations to execute on a client device. The client device determines the timing for performing the tests and other operations by managing the timing of polling requests to the device manager. If a device manager detects an event associated with a client device, the device manager generates an action record. At a time determined by the client device, the client device polls the device manager to identify the action record. The device manager transmits a polling response to the client device in response to the polling request. The client device executes code to perform the operations specified in the polling response. The code is stored in the client device prior to receiving the polling response.

Abstract: Techniques for determining cabling resources for a data center are disclosed. A system includes a data center model that includes a first device model entity, a second device model entity, and positional information corresponding to positions of the device model entities in a physical environment. Based on a physical linking rule that applies to the first device model entity and the second device model entity, and that defines one or more physical connections between the first device model entity and the second device model entity, the system generates one or more cabling entities that indicate one or more physical connections between physical devices represented by the device model entities. Based on the cabling entities and the positional information, the system outputs a set of one or more attributes for one or more physical cables that correspond to the cabling entities for use in a bill of materials.

Abstract: Security vulnerabilities in programming language source code generated by Language Models (LMs) are addressed through a two-part approach combining static analysis with LM-based remediation. In an implementation, a “base” LM is used for initial code generation, followed by a static analysis tool that identifies security vulnerabilities and a smaller “guardrail” LM (1-7 billion parameters) trained for generating security fixes. Unlike traditional approaches that regenerate entire code blocks, the implementation uses targeted fixing through special tokens that mark vulnerable code sections, allowing precise modifications while maintaining code context and structure. The implementation encompasses an iterative validation process to ensure fixes resolve identified vulnerabilities and can handle complex scenarios, including multi-line vulnerabilities and cross-file dependencies.

Abstract: Techniques for generating permissions policies for nested sets of software artifacts are disclosed. A system detects a selection of a target action and a target software artifact. The system identifies an entity associated with the selection. The system analyzes permissions metadata of the target software artifact to identify user-defined permission configuration rules specifying one or more additional software artifacts and one or more additional actions associated with the additional software artifacts. The system performs an iterative analysis of permissions metadata in the nested software artifacts to generate a composite permissions policy for the entity that combines the permissions specified in the user-defined permission configuration rules for the set of nested software artifacts.

Abstract: The present disclosure relates to methods and systems for protecting cloud resources. The methods and systems may use a virtual gatekeeper resource to enforce secure access controls to cloud resources for a list of privileged operations. The cloud resources and the virtual gatekeeper resource may be in different security domains within a cloud computing system and the cloud resources may be linked to the virtual gatekeeper resource. A request may be sent to perform a privileged operation on the cloud resource. Access may be provided to the virtual gatekeeper resource in response to approval of the request and the access to the virtual gatekeeper resource may be used to perform the privileged operation on the cloud resource.

Abstract: Techniques for managing the implementation of application-code scanning processes are disclosed. A system scans application code by analyzing metadata associated with the application code to identify a set of data needed to scan the application code with a scanning application. Based on the information obtained from the application metadata, the system identifies extraction processes that are needed to obtain the set of data. The system applies a set of one or more application-code scanners by implementing the extraction processes. The system presents in a graphical user interface (GUI) a set of results from scanning operations.

Abstract: Techniques for using machine learning to generate skill hierarchies are disclosed. A system generates a graph that models a set of skills, including first and second nodes that represent first and second subsets of the skills. The system uses a model to determine that the first and second nodes are candidates to be connected by an edge in the graph, then uses a large language model to determine that the nodes are sufficiently similar to be connected by the edge. The system performs a minimum cut of a connected subgraph that includes the first and second nodes, to obtain first and second connected sub-subgraphs. The system uses generative artificial intelligence to determine (a) first and second node-level representatives for the first and second nodes and (b) a sub-subgraph-level representative skill. The system generates a skill hierarchy that includes the node-level and sub-subgraph-level representative skills.

Abstract: Techniques for compiling switch blocks are disclosed. One or more embodiments analyze and rewrite a set of pattern labels in a switch block in a compile-time process for determining whether the switch block is exhaustive. At compile-time, a system populates a set with case labels from a switch block. The system applies a set of rules to iteratively re-write pattern labels in the set and checks whether the re-written set, and hence the original switch block, is exhaustive. If the compiler determines that (a) the set does not appear to be exhaustive, and (b) the set includes patterns labels, then the compiler determines whether the set may be rewritten before further analysis. The compiler iteratively re-writes and re-analyzes the case label set for exhaustivity until the case label set is determined to be exhaustive or cannot be rewritten further.

Abstract: Techniques are described for transmitting metric data between tenancies. Metric data is gathered for resources within a customer tenancy of a multi-tenant environment. This metric data is sent to a service tenancy of the multi-tenant environment, where the service tenancy is separate from the customer tenancy. The metric data is validated and preprocessed within the service tenancy to make sure that all required fields (such as key-value pairs) are located within the metric data. The preprocessed metric data is then sent to a telemetry service for analysis.

Abstract: A computer program product, system, and computer implemented method for an approach to manage objects that tightly couple snapshots and clone management together. Specifically, the present disclosure provides for an approach where metadata for snapshots and clones of the extent can be managed using a single set of metadata that intermingles entries for any number of snapshots and clones together. This provides for a single place in which the storage system can access extent information for a given extent to determine what information corresponds to a request (e.g., create, read, update, delete requests). Additionally, the approaches provided herein further include support for variable sized allocation units and tombstones (that represent sparse data using metadata). In some embodiments, metadata and data referenced by corresponding metadata are stored on the same storage devices.

Abstract: Techniques for generating accurate responses to queries using multiple language models and source document validation are disclosed. The techniques include receiving a source document and a query that are then processed through multiple language models to generate various responses. These responses are analyzed to identify segments that are both relevant to the query and validated against the source document for accuracy. From these validated segments, a second subset is determined. The techniques then order these segments based on their relative positions within either the source document or the language model responses. A comprehensive response is generated using this ordered set of segments. This approach ensures that the final response is accurate, relevant, and maintains a logical flow based on the original document structure. The techniques combine multiple language mode outputs while maintaining source document fidelity through systematic validation and ordering.

Abstract: Techniques for identifying errors in cabled connections in a network are disclosed. A system determines that a new detected connection between a first device and a second device has been added to the network. The system measures a signal value associated with a signal characteristic for a signal transmitted between the first and second devices. The system compares the measured signal value to an expected signal value. When the measured signal value does not satisfy a criterion with respect to the expected signal value, the system outputs information indicating an error associated with the connection between the first device and the second device.

Abstract: Techniques for selectively isolating interconnectable modules in a coolant system are disclosed. A system directs a flow of coolant liquid through coolant pipe assembly racks mounted in a data center pod. The racks store heat-generating computing devices. The coolant pipe assembly may be configured as a loop-shaped structure. One side of the loop may supply coolant liquid to one row of racks, and another side of the loop may supply coolant liquid to another row of racks. The system closes a set of valves along a coolant pipe to isolate a set of target racks from the flow of coolant liquid. Due to the loop structure of the coolant pipe assembly, the system continues to direct coolant liquid to other racks in the same row of the loop-shaped structure as the isolated target racks.