Abstract: A data set can be provided that includes an input data element and one or more label data portion definitions that each identify a feature of interest within the input data element. A machine-learning model can generate model-identified portions definitions that identify predicted feature of interests within the input data element. At least one false negative (where a feature of interest is identified without a corresponding predicted feature of interest) and at least one false positive (where a predicted feature of interest is identified without a corresponding feature of interest) can be a identified. A class-disparate loss function can be provided that is configured to penalize false negatives more than at least some false positives. A loss can be calculated using the class-disparate loss function. A set of parameter values of the machine-learning model can be determined based on the loss.

Abstract: A method including extracting data from disparate data sources. The data includes data pairs including a corresponding data point and a corresponding time associated with the corresponding data point. The method also includes extracting insights from the data at least by identifying a trend in the data pairs. The method also includes forming a model vector including the insights and an additional attribute to the insights. The additional attribute characterizes the insights. The additional attribute includes at least user feedback including a user ranking of a ranked subset of the insights from a user. The method also includes inputting the model vector into a trained insight machine learning model to obtain a predicted ranking of the insights. The method also includes selecting, based on the predicted user ranking, a pre-determined number of insights to form predicted relevant insights. The method also includes reporting the predicted relevant insights.

Abstract: A computing system and method can be used to implement a version of federated learning (FL) that incorporates adaptivity (e.g., leverages an adaptive learning rate). In particular, the present disclosure provides a general optimization framework in which (1) clients perform multiple epochs of training using a client optimizer to minimize loss on their local data and (2) a server system updates its global model by applying a gradient-based server optimizer to the average of the clients' model updates. This framework can seamlessly incorporate adaptivity by using adaptive optimizers as client and/or server optimizers. Building upon this general framework, the present disclosure also provides example specific adaptive optimization techniques for FL which use per-coordinate methods as server optimizers. By focusing on adaptive server optimization, the use of adaptive learning rates is enabled without increase in client storage or communication costs and compatibility with cross-device FL can be ensured.

Abstract: Aspects relate to system and methods for determining a user specific mission operational performance, using machine-learning processes.

Abstract: Embodiments described herein provide visual-and-language (V+L) systems and methods for learning vision and language representations. Specifically, a method may comprise receiving a training dataset comprising a plurality of image samples and a plurality of text samples; encoding the plurality of image samples into a plurality of encoded image samples and the plurality of text samples into a plurality of encoded text samples; computing a first loss objective based on the plurality of encoded image samples and the plurality of encoded text samples; encoding a first subset of the plurality of encoded image samples and a second subset of the plurality of encoded text samples into a plurality of encoded image-text samples; computing a second loss objective based on the plurality of encoded image-text samples; and updating the V+L model based at least in part on the first loss objective and the second loss objective.

Abstract: One embodiment of the present invention sets forth a technique for curating a data sample set. The technique includes determining one or more data sampling criteria based on a sampling objective for a data sample set associated with the machine learning model. The technique also includes selecting, from a set of unlabeled data samples, at least one data sample to be labeled and added to a data sample set associated with the machine learning model based on the one or more data sampling criteria. The technique also includes, for each selected data sample, supplementing the data sample set with the selected data sample and at least one association with a label.

Abstract: A computer-implemented method is provided for policy evaluation. In the method, the utility of the given decision-making policy is estimated based on a dataset of state-action-reward-state tuples, a set of candidate bootstrapping estimators of the fitted Q-evaluation (FQE) algorithm, and a criterion function. The method automatically selects the best bootstrapping estimator from the candidates based on the criterion function and, when the criterion function is appropriately designed, it produces a good policy-value estimate such that the estimation error is small (below a threshold).

Abstract: A platooning control device includes: a learning device configured to perform reinforcement learning on the basis of image information and a feedback signal and to control a pertinent vehicle so as to follow a traveling trajectory of a front vehicle according to a result of the reinforcement learning; and a compensation determination unit configured to receive a coordinate of a control point regarding the traveling trajectory of the front vehicle from the front vehicle and to compare a coordinate of the pertinent vehicle with the coordinate of the control point, thereby generating the feedback signal.

Abstract: A system for determining whether inconsistencies exist in an entity's shared databases using a machine learning model. The system includes a repository having a plurality of databases that store data and information in a format accessible to users, and a back-end server operatively coupled to the repository and being responsive to the data and information from all of the databases. The back-end server includes a processor for processing the data and information, a communications interface communicatively coupled to the processor, and a memory device storing data and executable code. The code causes the processor to collect data and information from the databases, store the collected data and information in the memory device, process the stored data and information through the machine learning model to determine whether inconsistencies in the data exist in the databases, and transmit a communication on the interface identifying whether inconsistencies do exist in the databases.

Abstract: In some examples, a node for a telecommunication network includes a neural-network-based receiver for uplink communications. The node is configured to modify the neural-network-based receiver to generate a set of modified receiver frameworks defining respective different versions for the receiver, using each of the modified receiver frameworks, generate respective measures representing bits encoded by a signal received at the node, calculate a value representing a variance of the measures, and on the basis of the value, determine whether to select the signal received at the node for use as part of a training set of data for the neural-network-based receiver.

Abstract: An adaptive quantum signal processor (AQSP) includes a signal combiner, a physics station, a measurement system, a machine-learning engine and an output generator. The signal combiner combines incoming signals with control functions to yield recipe functions. For example, the recipe functions can be “shaking” functions used to change the wavefunctions of atoms entrained in an optical lattice. The recipe functions are applied to wavefunctions in initial wavefunction states causing the wavefunctions to transition to signal-impacted states. The measurement system measures the wavefunctions in their signal-impacted quantum states to yield wavefunction characterizations. The machine-learning engine updates control functions based on the wavefunction characterizations. The output generator outputs results based on the wavefunction characterizations and/or control function characterizations. In a matched-filter application, the outputs characterize (e.g., identify, classify, rate) the incoming signals.

Abstract: Techniques described herein relate to a method for managing a distributed multi-tiered computing (DMC) environment. The method includes obtaining, by a global controller, a DMC environment management request from a user, whereas the DMC environment management request is associated with scheduling an application in the DMC environment; and in response to obtaining the request: identifying target domains for tasks associated with the application based on the request; selecting scheduling policies for each target domain; obtaining fingerprints of previously provisioned applications associated with the request; performing case based reasoning using the fingerprints to generate scheduling packages for the target domains; sending the scheduling packages to local controllers associated with the target domains; obtaining application information from the local controllers; and providing the application information to the user.

Abstract: Methods, systems, and computer programs are presented for eliminating bias while training an ML model using training data that includes past experimental data. One method includes accessing experiment results, for A/B testing of a first model, that comprise information regarding engagement with a first set of items presented to users, each item being presented within an ordered list of results. A position bias is calculated for positions within the ordered list of results where the items were presented. A machine-learning program is trained to obtain a second model using a training set comprising values for features that include the calculated position bias. The method includes detecting a second set of items to be ranked for presentation to a first user, and calculates, using the second model, a relevance score for the second set of items, which are ranked based on the respective relevance score and presented on a display.

Abstract: Systems and methods for operating a quantum processor. The methods comprise: receiving a reward matrix at the quantum processor, the reward matrix comprising a plurality of values that are in a given format and arranged in a plurality of rows and a plurality of columns; converting, by the quantum processor, the given format of the plurality of values to a qubit format; performing, by the quantum processor, subset summing operations to make a plurality of row selections based on different combinations of the values in the qubit format; using, by the quantum processor, the plurality of row selections to determine a normalized quantum probability for a selection of each row of the plurality of rows; making, by the quantum processor, a decision based on the normalized quantum probabilities; and causing, by the quantum processor, operations of an electronic device to be controlled or changed based on the decision.

Abstract: A systematic explainer is described herein, which comprises local, model-agnostic, surrogate ML model-based explanation techniques that faithfully explain predictions from any machine learning classifier or regressor. The systematic explainer systematically generates local data samples around a given target data sample, which improves on exhaustive or random data sample generation algorithms. Specifically, using principles of locality and approximation of local decision boundaries, techniques described herein identify a hypersphere (or data sample neighborhood) over which to train the surrogate ML model such that the surrogate ML model produces valuable, high-quality information explaining data samples in the neighborhood of the target data sample.

Abstract: A system and method for performing machine learning in a mobile computing device which is configured to be coupled with a cloud computing system is disclosed. The method may include activating, on the mobile computing device, a machine learning application, which accesses a local machine learning system including a local machine learning model, periodically updating the local machine learning system based upon updates for the local machine learning system received from a global machine learning system hosted by the cloud computing system, performing machine learning based on received training data, and periodically transmitting changes to the local machine learning system from the mobile computing device to the global machine learning system hosted by the cloud computing system.

Abstract: In some embodiments, a service platform that facilitates artificial intelligence model and data collection and collection may be provided. Input/output information derived from machine learning models may be obtained via the service platform. The input/output information may indicate (i) first items provided as input to at least one model of the machine learning models, (ii) first prediction outputs derived from the at least one model's processing of the first items, (iii) second items provided as input to at least another model of the machine learning models, (iv) second prediction outputs derived from the at least one other model's processing of the second items, and (v) other inputs and outputs. The input/output information may be provided via the service platform to update a first machine learning model. The first machine learning model may be updated based on the input/output information being provided as input to the first machine learning model.

Abstract: Disclosed examples include an automated online experimentation mechanism that can perform model selection from a large pool of models with a relatively small number of online experiments. The probability distribution of the metric of interest that contains the model uncertainty is derived from a Bayesian surrogate model trained using historical logs. Disclosed techniques can be applied to identify a superior model by sequentially selecting and deploying a list of models from the candidate set that balance exploration-exploitation.

Abstract: An AI-based platform for enabling intelligent orchestration and management of power and energy is disclosed. The platform includes a digital twin system having a digital twin of a mining environment. The digital twin includes at least one parameter that is detected by a sensor of the mining environment. In some disclosed embodiments, the at least one parameter is associated with one or more of an unmined portion of the mining environment a mining of materials from the mining environment, a smart container event involving a smart container associated with the mining environment, a physiological status of a miner associated with the mining environment, a transaction-related event associated with the mining environment, and a compliance of the mining environment with one or more contractual, regulatory, and/or legal policies.

Abstract: A data processing service facilitates the creation and processing of data processing pipelines that process data processing jobs defined with respect to a set of tasks in a sequence and with data dependencies associated with each separate task such that the output from one task is used as input for a subsequent task. In various embodiments, the set of tasks include at least one cleanroom task that is executed in a cleanroom station and at least one non-cleanroom task executed in an execution environment of a user where each task is configured to read one or more input datasets and transform the one or more input datasets into one or more output datasets.

Abstract: Aspects of the disclosure relate to intelligent bot performance tracking and analysis. A computing platform may receive a work queue of items to be processed using a bot. The computing platform may receive, in real-time with processing of the work queue using the bot, metadata associated with the work queue. Based on the metadata, the computing platform may assign, in real-time, a value metric associated with completion of each item in the work queue. Based on the assigned value metric, the computing platform may identify a robotic process automation cost associated with processing the work queue via the bot. The computing platform may compare, the robotic process automation cost to a cost to process the work queue via another operation, and determine a performance metric for the bot based on the comparison. The computing platform may dynamically generate and transmit, in real-time, an indication of the determined bot performance metric.

Abstract: A method of retrieving a document according to an embodiment of the present application includes: acquiring a user retrieval query; calculating a user inquiry vector in a unit of sentence from the user retrieval query and acquiring a first document candidate group based on similarity between the calculated user inquiry vector and an embedding vector of a document stored in a retrieval database; acquiring a second document candidate group based on similarity between a text included in the user retrieval query and a text of the document stored in the retrieval database; and determining a summarization target document based on the first document candidate group and the second document candidate group.

Abstract: A computing platform may be configured to (i) obtain an input dataset, (ii) construct a graph from the input dataset, (iii) for a given node within the constructed graph, generate a first type of embedding vector using a first embedding technique (e.g., a shallow embedding technique) and a second type of embedding vector using a second embedding technique that differs from the first embedding technique (e.g., a deep embedding technique), and (iv) use the first and second types of embedding vectors for the given node and a data science model to render a given prediction for the given node.

Abstract: A computer-implemented method for performing Learning from Demonstrations, particularly Imitation Learning, based on data associated with a first domain, particularly a source domain. The method includes: determining first data characterizing a demonstrator of the first domain, wherein particularly the first data characterizes sensor data of the demonstrator and/or sensor data of at least one spectator observing the demonstrator, determining first knowledge from the first domain based on the first data, transferring at least a part of the first knowledge to a second domain, particularly a target domain.

Abstract: There is provided a system and method of runtime examination of a semiconductor specimen. The method includes obtaining a runtime image representative of an inspection area of the specimen, the runtime image having a relatively low signal-to-noise ratio (SNR); and processing the runtime image using a machine learning (ML) model to obtain examination data specific for a given examination application, wherein the ML model is previously trained for the given examination application using one or more training samples, each training sample representative of a respective reference area sharing the same design pattern as the inspection area and comprising: a first training image of the respective reference area having a relatively low SNR; and label data indicative of ground truth in the respective reference area pertaining to the given examination application, the label data obtained by annotating a second training image of the respective reference area having a relatively high SNR.

Abstract: One or more computing devices, systems, and/or methods are provided. A machine learning model may be trained using a plurality of sets of information. One or more pruning operations may be performed in association with the training to generate a machine learning model with sparse vector representations associated with features of the plurality of sets of information. A request for content associated with a client device may be received. A set of features associated with the request for content may be determined. A plurality of positive signal probabilities associated with a plurality of content items may be determined using the machine learning model based upon one or more sparse vector representations, of the machine learning model, associated with the set of features. A content item may be selected from the plurality of content items for presentation via the client device based upon the plurality of positive signal probabilities.

Abstract: Systems and methods of computing classifications for and migrating digital content that includes accessing a digital content corpus within a source data storage system; in response to accessing the digital content corpus, for each distinct item of digital content of the plurality of distinct items of digital content: computing, via one or more digital content machine learning classification models, a content classification inference; identifying automated digital content handling tasks of a plurality of distinct digital content handling tasks based on the content classification inference; executing the automated content handling tasks identified for each distinct item of digital content, wherein executing the automated content handling tasks includes: designating a storage location within a target data storage system based on the in-migration content classification inference; and migrating a respective item of digital content from the source data storage system to the designated storage location within the ta

Abstract: A method for predicting a property associated with a product unit. The method may include: obtaining a plurality of data sets, wherein each of the plurality of data sets includes data associated with a spatial distribution of a parameter across the product unit; representing each of the plurality of data sets as a multidimensional object; obtaining a convolutional neural network model trained with previously obtained multidimensional objects and properties of previous product units; and applying the convolutional neural network model to the plurality of multidimensional objects representing the plurality of data sets, to predict the property associated with the product unit.

Abstract: In an approach, a processor receives device identification information corresponding to at least one device local to a location of a transaction. A processor receives notification of an infected user. A processor determines that the infected user is associated with the transaction. A processor identifies a second user from the device identification information. A processor sends a notification to the second user.

Abstract: An electronic device and a method of training a classification model for age-related macular degeneration (AMD) are provided. The method includes the following steps. Training data is obtained. A loss function vector corresponding to the training data is calculated based on a machine learning algorithm, in which the loss function vector includes a first loss function value corresponding to a first classification of AMD and a second loss function value corresponding to a second classification of AMD, the first classification corresponds to a first group, and the second classification corresponds to one of the first group and a second group. The first loss function value is updated according to the second loss function value and a group penalty weight in response to the second classification corresponding to the second group to generate an updated loss function vector. The classification model is trained according to the updated loss function vector.

Abstract: Systems and methods for intelligently clustering alerts and applying a multi-stage scoring approach to prioritize and effectively address cybersecurity alerts are disclosed. The multi-stage scoring approach may involve applying Time Series Frequency-Inverse Document Frequency (TSF-IDF) Scores that represent a novelty of a cluster and Confidence Scores that represent a measure of accuracy based on prior performance including true positives and other indicia of accuracy.

Abstract: Systems, methods, and apparatuses for providing dynamic, prioritized spectrum utilization management. The system includes at least one monitoring sensor, at least one data analysis engine, at least one application, a semantic engine, a programmable rules and policy editor, a tip and cue server, and/or a control panel. The tip and cue server is operable utilize the environmental awareness from the data processed by the at least one data analysis engine in combination with additional information to create actionable data.

Abstract: A method includes identifying, from online clustering data, an internet protocol (IP) pair. The method further includes determining, by a processing device during an offline process, that the IP pair is part of a botnet. The method further includes, in response to the determining, appending data associated with the botnet to the online clustering data.

Abstract: Aspects of the present disclosure provide systems, methods, and computer-readable storage media that support secure training of machine learning (ML) models that preserves privacy in untrusted environments using distributed executable file packages. The executable file packages may include files, libraries, scripts, and the like that enable a cloud service provider configured to provide ML model training based on non-encrypted data to also support homomorphic encryption of data and ML model training with one or more clients, particularly for a diagnosis prediction model trained using medical data. Because the training is based on encrypted client data, private client data such as patient medical data may be used to train the diagnosis prediction model without exposing the client data to the cloud service provider or others. Using homomorphic encryption enables training of the diagnosis prediction model using encrypted data without requiring decryption prior to training.

Abstract: A method and system of determining whether a stationary vehicle is a blocking vehicle to improve control of an autonomous vehicle. A perception engine may detect a stationary vehicle in an environment of the autonomous vehicle from sensor data received by the autonomous vehicle. Responsive to this detection, the perception engine may determine feature values of the environment of the vehicle from sensor data (e.g., features of the stationary vehicle, other object(s), the environment itself). The autonomous vehicle may input these feature values into a machine-learning model to determine a probability that the stationary vehicle is a blocking vehicle and use the probability to generate a trajectory to control motion of the autonomous vehicle.

Abstract: Systems and methods are directed to controlling components of a utility grid. The system can receive data samples including signals detected at one or more portions of a utility grid. The system can construct a matrix having a first dimension and a second dimension. The system can train a machine learning model based on the matrix to predict values for signals of the utility grid not provided in the matrix. The system can receive bounds for one or more input variables, constraints on one or more output variables, and a performance objective for the utility grid. The system can determine, based on the machine learning model and via an optimization technique, an adjustment to a component of the utility grid that satisfies the performance objective. The system can provide the adjustment to the component of the utility grid to satisfy the performance objective.

Abstract: A method for training a machine learning algorithm including uncertainties. The method includes: pre-training the algorithm based on initially collected data by a control unit in order to obtain an initial model, determining a set of channels, the data originating from channels contained in the set of channels being intended to be used for retraining the initial model, based on an established data level and on the respective influence, which the data originating from one of the channels have on uncertainties instantaneously contained in the initial model, transferring detected data originating from the individual channels of the set of channels to the control unit, and retraining of the initial model by the control unit based on the data transferred to the control unit.

Abstract: Systems and methods provide for deriving values for a set of features, for each session of a plurality of sessions that each comprise one or more actions performed by a given user on one or more webpages of a website. The systems and method further provide for generating an initial frustration score for each session of the plurality of sessions by analyzing the set of features for each session of the plurality of sessions using a first machine learning model trained to generate an initial frustration score based on values derived for a set of features for each session of a given set of sessions.

Abstract: The exemplary embodiments disclose a method, a computer program product, and a computer system for determining that one or more model pipelines satisfy one or more constraints. The exemplary embodiments may include detecting a user uploading data, one or more constraints, and one or more model pipelines, collecting the data, the one or more constraints, and the one or more model pipelines, and determining that one or more of the model pipelines satisfies all of the one or more constraints based on applying one or more algorithms to the collected data, constraints, and model pipelines.

Abstract: A big data analysis system may include a big data repository communicatively coupled to a data accumulation server and a predictive graph processing system. The data accumulation server may be configured to receive information from a plurality of data sources, the information corresponding to user interaction with one or more computing devices associated with an organization via a networked computing system, store the information received from the plurality of sources in the big data repository; and monitor the plurality of data sources to update the data stored in the big data repository. The predictive graph processing system is configured to receive information stored in the big data repository, transform the information received from the big data repository into a predictive graph data set based on a predictive model, and store the predictive graph data set to a visualization data repository.

Abstract: This disclosure relates to systems and methods for identifying fraudulent email addresses associated with an electronic payment service. In some implementations, a computing device receives an email with a prefix having a number of characters and characterized by a prefix length indicative of the number of characters in the prefix. The computing device identifies each of a number of bigrams is identified within the prefix, and determines a row and column distance for each bigram between two consecutive characters of the bigram as positioned on a keyboard. The computing device calculates a Euclidean distance between the two consecutive characters of the bigram based on the row and column distances, and determines a normalized distance based on the prefix length and an average of the Euclidean distances calculated for the number of bigrams in the prefix. The normalized distance is compared with a value to classify the email as suspicious or as not suspicious.

Abstract: A centralized skills management server, a computer-readable storage medium, and a computer-implemented method for skills inference are described herein. The method includes executing a web-based application on a remote computing system operated by a user associated with a tenant and extracting skills-related terms associated with the execution of the web-based application. The method includes interfacing with the global skills graph via an API and importing standardized skill tags relating to the extracted skills-related terms.

Abstract: According to one embodiment, in nth (n is a natural number) processing, a first node calculates a first gradient to update a first weight and a second node calculates a second gradient to update the first weight. In mth (m is a natural number) processing, a third node calculates a third gradient to update a third weight and a fourth node calculates a fourth gradient to update the third weight. If the calculation by the first and second nodes is faster than the calculation by the third and fourth nodes, in n+1th processing, a second weight updated from the first weight is further updated using the first and second gradients, and, in m+1th processing, a fourth weight updated from the third weight is further updated using the first to fourth gradients.

Abstract: Systems, methods, devices and non-transitory, computer-readable storage mediums are disclosed for a wearable multimedia device and cloud computing platform with an application ecosystem for processing multimedia data captured by the wearable multimedia device. In an embodiment, a method comprises: receiving, by one or more processors of a cloud computing platform, context data from a wearable multimedia device, the wearable multimedia device including at least one data capture device for capturing the context data; creating a data processing pipeline with one or more applications based on one or more characteristics of the context data and a user request; processing the context data through the data processing pipeline; and sending output of the data processing pipeline to the wearable multimedia device or other device for presentation of the output.

Abstract: Systems and methods for searching subsets of a search space. The system includes a memory with programmable instructions for searching a search space stored thereon, and processor for executing the programmable instructions. A user enters a search query, for example, via the user interface. The system receives a search command including the search query. The system performs a first search of a first subset of the of the search space using the search query. The results from the first search are presented to the user. The system receives a second, supplemental search command from the user, and responsive to the second, supplemental search command, performs a second search, using the same search query, of a second subset of the search space.

Abstract: Processing within a computing environment is facilitated using a corpus processing system to assess and enhance quality of a corpus of unstructured documents for a specified task. The processing includes referencing, by a corpus processing engine, the corpus of unstructured documents to obtain unstructured document data, and applying, by a corpus quality metrics engine, a set of quality metrics to the document data to obtain a set of quality metric scores. Further, the process includes automatically selecting, by a quality metric selection engine, a subset of task-relevant quality metrics using the quality metric scores and the specified task, and automatically transforming, at least in part, multiple documents of the corpus to remediate one or more identified issues with the documents. The automatically transforming results in remediated documents tuned for the specified task, which are provided for the specified task to be performed.

Abstract: The technology relates to machine responses to anomalies detected using machine learning based anomaly detection. In particular, to receiving evaluations of production events, prepared using activity models constructed on per-tenant and per-user basis using an online streaming machine learner that transforms an unsupervised learning problem into a supervised learning problem by fixing a target label and learning a regressor without a constant or intercept. Further, to responding to detected anomalies in near real-time streams of security-related events of tenants, the anomalies detected by transforming the events in categorized features and requiring a loss function analyzer to correlate, essentially through an origin, the categorized features with a target feature artificially labeled as a constant.

Abstract: Virtual platform system for use in a cloud-based system, comprising: a virtual platform simulator configured to represent in software a physical remote client device and to have this representation interact with a virtual platform application; a process virtual machine configured to execute program instructions of the virtual platform application and comprising a code morpher component for transforming the program instructions of the virtual platform application into native program instructions for execution on a physical host machine of the cloud-based system; and interception components for capturing transactions from the virtual platform simulator and the process virtual machine. The transactions are related to the execution of the program instructions of the virtual platform application.

Abstract: Implementing artificial intelligence, specifically, machine learning techniques to identify malicious emails and, in response, identifying and conducting actions, including reporting the malicious emails to identified internal and/or external entities and preventing the malicious emails from being delivered to email client mailboxes. The machine learning techniques rely on malicious email patterns identified, at least, from previously identified malicious emails and data resulting from continuously crawling the Web and threat intelligence sources. Further, the email clients may be configured to include an add-on feature in which the user can provide a single input to report the email as being suspicious, which results in further analysis to determine whether the email is, in fact, a malicious email.

Abstract: Techniques are disclosed relating to the execution of machine learning models on client devices, particularly in the context of transaction risk evaluation. This reduces computational burden on server systems. In various embodiments, a server system may receive, from a client device, a request to perform a first operation and select a first machine learning model, from a set of machine learning models, to send to the client device. In some embodiments the first machine learning model is executable, by the client device, to generate model output data for the first operation based on one or more encrypted input data values that are encrypted with a cryptographic key inaccessible to the client device. The server system may send the first machine learning model to the client device and then receive, from the client device, a response message that indicates whether the first operation is authorized based on the model output data.