Machine learning

MACHINE LEARNING-BASED MODIFICATION OF IMAGE CONTENT

Resumen de: US20260038163A1

Aspects of the present disclosure involve a system comprising a computer-readable storage medium storing a program and method for modifying a captured image. The program and method provide for displaying, by a messaging application, an image captured by a device camera; providing, by the messaging application, a user interface for selecting from among a plurality of content modifiers to modify the image, the plurality of content modifiers including a first content modifier corresponding to a machine learning model trained with a plurality of image pairs, each image pair including a first image and a second image corresponding to a modified version of the first image; receiving user selection of the first content modifier from among the plurality of content modifiers; determining, in response to receiving the user selection, a modified version of the image based on output from the machine learning model; and displaying the modified version of the image.

EXECUTION AND SEMANTIC ERROR CORRECTION CAPABILITIES FOR NATURAL LANGUAGE TO LOGICAL FORM MODEL

Resumen de: WO2026030330A1

Techniques are disclosed herein for providing and using a natural language to logical form model having execution and sematic error correction capabilities. In one aspect, a method is disclosed that includes: accessing a set of training examples and generating a set of error correction training examples via an iterative process performed for each training example. The iterative process includes generating an inferred logical form, executing the inferred logical form on a database, when executing the inferred logical form on the database fails, obtaining an execution error message corresponding to the failure, and recording the inferred logical form and the execution error message as part of an execution error example, and populating an error correction prompt template with the execution error example to generate an error correction training example. A machine learning model may then be trained with at least the set of error correction training examples.

INTERACTIVE INTERFACE TASK AUTOMATION UTILIZING GENERATIVE ARTIFICIAL INTELLIGENCE (AI) ACTION MODELS IMPROVED WITH RETRIEVAL-AUGMENTED GENERATION (RAG)

Resumen de: US20260037318A1

This disclosure describes a framework for performing user-requested tasks automatically across an interactive interface using various types of machine learning models. Specifically, this disclosure outlines and describes a task execution system that utilizes a generative artificial intelligence (AI) action model and retrieval-augmented generation (RAG) to complete user-requested actions across an interactive interface. The task execution system solves many of the current limitations of LAMs by using a generative AI action model to determine a session plan, which includes a set of actions for accomplishing stages of the actionable task across the interactive interface, obtaining visual context information of each interactive interface segment, integrates RAG results to improve the accuracy of both the session plan and individual actions, and self-corrects when faced with unexpected obstacles.

AUTOMATED AI/ML EVENT MANAGEMENT SYSTEM

Resumen de: US20260040099A1

Aspects of the subject disclosure may include, for example, receiving, from a machine learning model, information about an event causing a service degradation in a cellular network, wherein the event is external to the cellular network, determining one or more event categories associated with the event causing the service degradation, determining, based on the one or more event categories, likely affected customers, the likely affected customers being likely to experience the service degradation, determining, by the machine learning model, proper resources for resolution of the service degradation, wherein the determining proper resources is based on the one or more event categories, and dispatching the proper resources for resolution of the service degradation. Other embodiments are disclosed.

Detection of User Interface Imitation

Resumen de: US20260039693A1

Techniques are disclosed relating to generating trained machine learning modules to identify whether user interfaces accessed by a computing device match user interfaces associated with a set of Internet domain names. A server computer system receives a set of Internet domain names and generates screenshots for user interfaces associated with the set of Internet domain names. The server computer system then trains machine learning modules that are customized for the set of Internet domain names using the screenshots. The server then transmits the machine learning modules to the computing device, where the machine learning modules are usable by an application executing on the computing device to identify whether a user interface accessed by the device matches a user interface associated with the set of Internet domain names. Such techniques may advantageously allow servers to identify whether user interfaces are suspicious without introducing latency and increased page load times.

Computer Systems and Methods for Generating Predictive Change Events

Resumen de: US20260037890A1

Based on receiving data defining a new data item for a construction project corresponding to a particular category of data items, a computing system (1) automatically: (i) predicts that a change event for the construction project is needed by inputting the new data item into a first machine learning model trained to predict a need for a change event from data items corresponding to certain categories of data items, including the particular category of the new data item, (ii) determines initial recommended data for the predicted change event, and (iii) determines additional data for the predicted change event corresponding to a particular class of additional data by inputting the initial recommended data for the predicted change event into a second machine learning model trained to predict one or more classes of additional data for a change event, and (2) automatically create a data item representing the predicted change event.

Contrastive Explanations For Machine Learning Forecasting Models

Resumen de: US20260037842A1

A Contrastive Forecasting Explanation (CFE) tool and technique provides a model-agnostic approach to forecasting explanation. The CFE tool uses an ML-based surrogate forecaster as a surrogate model. The surrogate forecaster includes a time series preprocessor, a simple concept generator, and an ML forecaster. The subsequent interpretation of the predictions of the time series forecaster is based on the behavior of the surrogate forecaster. The CFE tool interprets time series forecasts by identifying the specific temporal concepts impacting predictions and thus generates clear and reliable explanations regardless of model type. The simple concepts and predictions generated by the surrogate model are input into a perturbation-based explainer to produce feature attributions from the surrogate model. An attribution postprocessor aggregates the attributions into more coherent concepts to present a coherent, concise, and interpretable explanation.

DATA AGGREGATION AND MODEL TRAINING BASED ON SPARSE DATASETS

Resumen de: US20260037871A1

A system may access a set of training data and determine a timeframe associated with a positively labeled data item of the training data. A system may generate at least two new positively labeled data items based on the positively labeled data item to generate augmented training data. A system may train a machine learning model by applying the augmented training data as input to a machine learning model, and modifying a weight of the machine learning model.

QUANTUM-SECURE MULTIPARTY DEEP LEARNING

Resumen de: WO2026030526A1

Quantum-secure, multiparty computation enables the joint evaluation of multivariate functions across distributed users while ensuring the privacy of their local inputs. It uses a linear algebra engine that leverages the quantum nature of light for information-theoretically secure multiparty inference using telecommunication components. This linear algebra engine can perform deep learning inference with rigorous upper bounds on the information leakage of both the deep learning model weights and the client's data, enabling double-blind operations. Applied to the MNIST classification task, it performs with classification accuracies exceeding 95% and a leakage of less than 0.1 bit per weight and data symbols. This leakage is an order of magnitude below the minimum bit precision for accurate deep learning using state-of-the- art quantization techniques. Our quantum-secure, multiparty computation lays the foundation for practical quantum-secure computation and unlocks secure cloud deep learning.

EXECUTION AND SEMANTIC ERROR CORRECTION CAPABILITIES FOR NATURAL LANGUAGE TO LOGICAL FORM MODEL

Resumen de: US20260037505A1

Techniques are disclosed herein for providing and using a natural language to logical form model having execution and sematic error correction capabilities. In one aspect, a method is disclosed that includes: accessing a set of training examples and generating a set of error correction training examples via an iterative process performed for each training example. The iterative process includes generating an inferred logical form, executing the inferred logical form on a database, when executing the inferred logical form on the database fails, obtaining an execution error message corresponding to the failure, and recording the inferred logical form and the execution error message as part of an execution error example, and populating an error correction prompt template with the execution error example to generate an error correction training example. A machine learning model may then be trained with at least the set of error correction training examples.

DATA AGGREGATION AND MODEL TRAINING BASED ON SPARSE DATASETS

Resumen de: WO2026030336A1

A system may access a set of training data and determine a timeframe associated with a positively labeled data item of the training data. A system may generate at least two new positively labeled data items based on the positively labeled data item to generate augmented training data. A system may train a machine learning model by applying the augmented training data as input to a machine learning model, and modifying a weight of the machine learning model.

INTERACTIVE INTERFACE TASK AUTOMATION UTILIZING GENERATIVE ARTIFICIAL INTELLIGENCE (AI) ACTION MODELS IMPROVED WITH RETRIEVAL-AUGMENTED GENERATION (RAG)

Resumen de: WO2026029823A1

This disclosure describes a framework for performing user-requested tasks automatically across an interactive interface using various types of machine learning models. Specifically, this disclosure outlines and describes a task execution system that utilizes a generative artificial intelligence (AI) action model and retrieval-augmented generation (RAG) to complete user-requested actions across an interactive interface. The task execution system solves many of the current limitations of LAMs by using a generative AI action model to determine a session plan, which includes a set of actions for accomplishing stages of the actionable task across the interactive interface, obtaining visual context information of each interactive interface segment, integrates RAG results to improve the accuracy of both the session plan and individual actions, and self-corrects when faced with unexpected obstacles.

COMPUTER SYSTEM AND METHOD FOR CLASSIFYING ASSETS IN AUTOMATED AND INDUSTRIAL CONTROL SYSTEMS

Resumen de: EP4687049A1

Classifying one or more assets in an automated and industrial control system (AIC) according to a classification standard. In a computer monitoring tool, a classification query is received for an asset managed by the AIC. Responsive to this classification query, the computer monitor tool retrieves a listing of candidate ontology classes for the queried asset utilizing information received from a semantic data model of known assets. The computer monitor tool then captures, preferably from a database coupled to the AIC, certain classification attribute variables associated with the queried asset. Additionally, the computer monitor tool receives user information describing certain building information associated with the queried asset. The computer monitor tool then generates a computer query configured for requesting results from a machine learning (ML) algorithm indicative of one or more classification standards for the queried asset.

ENRICHMENT AND RECONCILIATION OF FINANCIAL TRANSACTION DATA

Resumen de: WO2026022822A1

A system is configured to provide data pertaining to a record, performing the following method: (a) obtain, from a first source(s), a first record indicative of an actual financial transaction, paid via the first source; (b) perform enrichment on the first record, thereby determining: a counterparty and/or a financial classification category associated with the first record. The enrichment utilizes machine learning model(s) trained to identify correspondence, of first records indicative of actual financial transactions associated with a business entity, to second data. The second data are obtained from second source(s), distinct from the first source(s); (c) derive an enriched first record, based on the enrichment; and (d) provide the enriched first record.

CAPABILITIES AND SAFE PLUGINS

Resumen de: US20260030039A1

Disclosed are methods for managing execution of plugins of a machine-learning based system. A plugin configuration defines inputs required by the plugin and capabilities provided by the plugin. Capabilities describe the plugin’s functionality, such as how the plugin affects the response, what type of content the plugin generates, etc. In some configurations, when responding to a prompt, a collection of relevant plugins is identified. Configurations of these plugins may be analyzed to optimize execution, including determining optimal execution order or enabling parallel execution. Plugin configurations may also be analyzed to improve security by conditionally preventing one plugin from accessing the output of another. Plugin configurations may also be used to inform a client what plugins will run and what results they may yield. This enables the client to optimize and streamline how the response is displayed.

MACHINE LEARNING TECHNIQUES FOR DISCOVERING KEYS IN RELATIONAL DATASETS

Resumen de: AU2024318556A1

Techniques for discovering primary, unique, and/or foreign keys for relational datasets are described. The techniques include profiling the relational datasets to obtain respective data profiles; identifying one or more primary key candidates for a first relational dataset using a first data profile of the first relational dataset and a first trained machine learning model; identifying one or more foreign key proposals for a second relational dataset using the one or more primary key candidates by performing a subset analysis of the second relational dataset with respect to the first relational dataset; identifying one or more foreign key candidates for the second relational dataset using the first data profile, a second data profile of the second relational dataset, and a second trained machine learning model different from the first trained machine learning model; and outputting the at primary key candidate(s) and the foreign key candidate(s).

Information Processing Method, Program, and Information Processing Device

Resumen de: US20260030319A1

Provided is an information processing method, etc. that assists a user in interpreting behavior of a generated machine learning model. In the information processing method, a computer executes processing of recording a plurality of sets of an explanatory data vector xn input to an existing machine learning model (21) and an objective data vector yn output from the machine learning model (21) in association with each other, calculating an interpretation matrix A_dagger which is a vector product of an explanatory matrix X in which a plurality of sets of the explanatory data vector xn is arranged and a generalized inverse matrix of an objective matrix Y in which the objective data vector yn is arranged in an order corresponding to the explanatory data vector X, and outputting a chart (41, 42, and 43) related to the interpretation matrix A_dagger.

COMPUTER SYSTEM AND METHOD FOR CLASSIFYING ASSETS IN AUTOMATED AND INDUSTRIAL CONTROL SYSTEMS

Resumen de: US20260029780A1

Classifying one or more assets in an automated and industrial control system (AIC) according to a classification standard. In a computer monitoring tool, a classification query is received for an asset managed by the AIC. Responsive to this classification query, the computer monitor tool retrieves a listing of candidate ontology classes for the queried asset utilizing information received from a semantic data model of known assets. The computer monitor tool then captures, preferably from a database coupled to the AIC, certain classification attribute variables associated with the queried asset. Additionally, the computer monitor tool receives user information describing certain building information associated with the queried asset. The computer monitor tool then generates a computer query configured for requesting results from a machine learning (ML) algorithm indicative of one or more classification standards for the queried asset.

MACHINE LEARNING TECHNIQUES FOR GENERATING VISUALIZATION RECOMMENDATIONS

Resumen de: US20260030528A1

A visualization recommendation system generates recommendation scores for multiple visualizations that combine data attributes of a dataset with visualization configurations. The visualization recommendation system maps meta-features of the dataset to a meta-feature space and configuration attributes of the visualization configurations to a configuration space. The visualization recommendation system generates meta-feature vectors that describe the mapped meta-features, and generates configuration attribute sets that describe the attributes of the visualization configurations. The visualization recommendation system applies multiple scoring models to the meta-feature vectors and configuration attribute sets, including a wide scoring model and a deep scoring model. In some cases, the visualization recommendation system trains the multiple scoring models using the meta-feature vectors and configuration attribute sets.

SYSTEMS AND METHODS FOR PERSONALIZED GUIDANCE

Resumen de: US20260030516A1

Described are systems and method for personalized search results, including a memory storing instructions, a trained machine learning model, and a processor operatively connected to the memory and configured to execute the instructions to perform operations, including receiving the sequence of search queries from a user device associated with a user, predicting the likely next search query from the user by inputting the received sequence of search queries into the trained machine learning model, generating predicted search results by applying the likely next search query, generating the personalized search results by appending the predicted search results to search results from a most recent query of the sequence of queries from the user, and causing the user device to display the personalized search results.

CAPABILITIES AND SAFE PLUGINS

Resumen de: WO2026024342A1

Disclosed are methods for managing execution of plugins of a machine-learning based system. A plugin configuration defines inputs required by the plugin and capabilities provided by the plugin. Capabilities describe the plugin's functionality, such as how the plugin affects the response, what type of content the plugin generates, etc. In some configurations, when responding to a prompt, a collection of relevant plugins is identified. Configurations of these plugins may be analyzed to optimize execution, including determining optimal execution order or enabling parallel execution. Plugin configurations may also be analyzed to improve security by conditionally preventing one plugin from accessing the output of another. Plugin configurations may also be used to inform a client what plugins will run and what results they may yield. This enables the client to optimize and streamline how the response is displayed.

INFORMATION PROCESSING DEVICE, INFORMATION PROCESSING METHOD, SPAD IMAGING DEVICE, AI MODEL GENERATION METHOD, AI MODEL GENERATION DEVICE, AND INFERENCE PROCESSING DEVICE

Resumen de: EP4686214A1

An information processing device according to the present technology includes a calculation unit that calculates a lower limit value of a length of a count period for making likelihood information equal to or greater than a threshold on the basis of an inference result obtained by inputting a plurality of SPAD images obtained by an SPAD sensor to an AI model obtained by machine learning, the SPAD images having different count values by varying lengths of the count period in which photon counting is performed for each pixel, and likelihood information of the inference result.

SYSTEM AND METHOD FOR DYNAMIC MULTI-PARTY VERIFICATION OF GENERATIVE ARITIFICIAL INTELLIGENCE SYSTEMS

Resumen de: US20260025388A1

A model verification system and associated method for employing a multi-party verification technique to verify machine learning models and generative AI systems. The models and associated systems can be deployed in an enterprise and require verification to ensure that cohorts are properly verifying the models and systems and evaluation to ensure that the models and systems operate responsibly and achieve intended outcomes. A dynamic, multi-stakeholder blinded verification process can be employed for the continuous verification and evaluation of machine learning models and the systems that use them. This helps promote unbiased, reproducible verification, evaluation and assessments by preventing potential biases from cohorts form part of the verification process.

SAFETY NET ENGINE FOR MACHINE LEARNING-BASED NETWORK AUTOMATION

Resumen de: US20260025327A1

In one embodiment, a device obtains data regarding routing decisions made by a machine learning-based predictive routing engine for a network. The device determines, based on the data regarding the routing decisions, a behavior of the machine learning-based predictive routing engine. The device compares the behavior of the machine learning-based predictive routing engine to a behavioral policy for the machine learning-based predictive routing engine. The device adjusts operation of the machine learning-based predictive routing engine, when the behavior of the machine learning-based predictive routing engine violates the behavioral policy.

PREDICTION SELECTION FOR ITEM IDENTIFIERS USING EFFICIENT SELECTION ALGORITHM

Nº publicación: WO2026019632A1 22/01/2026

Solicitante:

MAPLEBEAR INC [US]
MAPLEBEAR INC

Resumen de: WO2026019632A1

A smart system, such as a smart shopping cart system, uses an efficient selection algorithm to select an item identifier prediction for an item. The smart cart system uses a set of machine-learning models to generate identifier predictions based on images. To select an item identifier, the smart system applies an efficient selection algorithm to the predictions from the machine-learning models. An efficient selection algorithm is an algorithm that requires minimal computational resources to perform. For example, the efficient selection algorithm may be a simple majority algorithm that selects the identifier prediction generated by a majority of the models or a weighted voting algorithm where each model's vote is weighted by some metric. The smart system applies the efficient selection algorithm to select an item identifier prediction from the ones generated by the models. The smart system may display content related to the item associated with the item identifier prediction.