Alerta

METHOD FOR IMPROVING ACCURACY OF MACHINE LEARNING MODELS

Resumen de: GB2642421A

Method for training a neuro-symbolic machine learning model, comprising: for each image depicting at least two objects of a training dataset: inputting the image into a neural module 102, 200, 202 to obtain bounding boxes and features therein (digit); inputting each bounding box and object feature into a symbolic module (106, Fig.1; rest of Fig.2) to obtain a plurality of possible labels i.e. partial labels 212 and possible relationships 210 as a new partially-labelled training dataset; and training the neuro-symbolic model (neural module and the symbolic module) by calculating a loss from a ground truth label for the image. The symbolic module may use a set of logical rules to constrain the labels and explanations (R1-R5, Fig.7). The trained neuro-symbolic model may generate a scene graph, perform action recognition, perform visual question answering (Fig.4) or control an autonomous or semi-autonomous electronic device. The electronic device may be a moveable robot or a wearable augmented reality device.

METHOD FOR DETERMINING A CONTRIBUTION OF A FEATURE COMBINATION TO AN OUTPUT OF A MACHINE LEARNING MODEL

Resumen de: EP4804085A1

A computer-implemented method, wherein the method determining a contribution of, optionally pairwise or higher-order, combinations of features between at least two application feature sets to an application output of a trained machine learning model, each combination of features comprising at least one feature of a first application feature set and at least one feature of a second application feature set.

SYSTEMS AND METHODS FOR DYNAMICALLY UPDATING MODELS USING MACHINE LEARNING

Resumen de: US20250148482A1

0000 A computing system for detecting patterns in data is provided. The computing system includes a model engine configured to receive an initial dataset, and segment the initial dataset into a plurality of subsets. The model engine is further configured to assign a weight to each subset based at least in part on an age of the subset, train a machine learning model on each subset separately in accordance with the assigned weighting for that subset. The model engine is further configured to receive a candidate dataset, analyze the candidate dataset using the trained machine learning model, and assign a score to the candidate dataset based on the analysis. The computing system further includes a rules engine configured to receive the candidate dataset and the corresponding score from the model engine, and generate and output, based at least in part on the score, a decision regarding the candidate dataset.

RISK ASSESSMENT OF ROTOR ANGLE INSTABILITY IN A POWER NETWORK

Resumen de: EP4804364A1

Rotor angle instability is a key criterion of dynamic stability in power networks. State-of-the-art machine-learning approaches are difficult to scale and have limited inputs with which to make predictions as to rotor angle instability. Accordingly, disclosed embodiments utilize a machine-learning model that is applied to bus voltage angles, which are local quantities available at every bus in the power network, to derive a prediction of the risk of rotor angle stability in the power network. These predictions may be biased in order to avoid false negatives. The machine-learning model may be a message-passing neural network. The resulting predictor is capable of quickly and reliably flagging potential rotor instability within a power network.

MITIGATING CYBER-THREATS IN A NETWORK

Resumen de: WO2026180318A1

A method of mitigating cyber-threats in a target network comprising a plurality of nodes, the method comprising: deploying a decentralised multi-agent model to the target network, wherein the decentralised multi-agent model comprises a plurality of local models; and at each of the plurality of nodes: monitoring a local region of the network to obtain local network information; and using the local model to predict threat mitigation actions, based on the local network information The decentralised multi-agent model may be a machine learning model trained using reinforcement learning wherein, for each of one or more training networks: the local models are deployed in respective nodes of the training network; and a trainer system iteratively evaluates a performance of the multi-agent model and adjusts the local models.

SCALABLE DETECTION OF DECENTRALIZED GENETIC BIOMARKERS IN HUMAN MICROBIAL DATA TO PERSONALIZE MEDICAL TREATMENT

Resumen de: WO2026181071A1

Detecting biomarkers in human microbiome DNA to predict a biological condition and administer its treatment. A microbiome network may be generated comprising nodes representing microbiome DNA sequences and edges representing a co-occurrence of each pair of microbiome DNA sequences in a same DNA sample or sub-length. Nodes may be bundled into distinct groups based on the node's degree quantifying a number of its edges indicating a number of unique microbiome DNA sequences that co¬ occur with the microbiome DNA sequence represented by the node in the same DNA sample or sub-length. Groups of bundled microbiome DNA sequences may be validated having an internal connectivity that satisfies an anomaly condition indicating the group's sequences co-occur with a probability that is unlikely randomly statistical, e.g., deviating from a power law distribution. A machine learning model may be trained with the validated groups to predict a biological condition correlated therewith to administer its treatment.

USING MACHINE LEARNING TECHNIQUES TO GENERATE SPATIO-TEMPORAL PREDICTIONS RELEVANT TO NOVEL GEOGRAPHIC AREAS

Resumen de: WO2026178648A1

In some embodiments, a computer-implemented method is provided. A computing system ingests historical data for a first geographic area and causal priors from a Bayesian model to create a knowledge graph. The computing system trains a spatial-temporal model to represent relationships in the knowledge graph over time and space. The computing system trains, using the knowledge graph and the spatial-temporal model, an imputation model to generate latent representations usable to infer missing data and to represent causal relationships. The computing system trains a generative model to generate counterfactual scenarios based on latent representations generated by the imputation model.

INTELLIGENT PARKING SPACE ALLOCATION SYSTEM AND METHOD WITH DYNAMIC MATCHING, MACHINE LEARNING AND BLOCKCHAIN INTEGRATION

Resumen de: WO2026180016A1

The invention relates to a computer-implemented method and system for dynamically allocating and coordinating parking spaces by means of real-time synchronisation between a departing driver and a driver searching for a parking space. In order to determine a precise departure time, the system uses sensor fusion comprising GNSS data, inertial sensor technology, and OBD-II vehicle data, as well as activity recognition. A hybrid machine-learning model comprising LSTM networks and random-forest regressors predicts the handover probability and generates an optimised assignment of the participants. Transaction security is ensured by a blockchain-based smart contract (preferably a layer-2 solution) which validates arrival, departure and payment by means of zero-knowledge proofs or a cryptographic handshake. By integrating predictive analytics and decentralised validation, traffic caused by drivers searching for parking is proactively reduced and utilisation of parking areas is optimised.

ADVERSARIAL TRAINING OF A MODEL FOR QUESTION ANSWERING

Resumen de: US20260260122A1

Embodiments of the present disclosure provide a solution for adversarial model training. A method includes: generating a prompt input using an adversarial machine learning model; providing the prompt input to a target machine learning model, to generate a response to the prompt input; determining a first reward score for the response with respect to the prompt input; and fine-tuning the target machine learning model according to a first optimization objective, the first optimization objective being configured to increase or maximize the first reward score for the target machine learning model.

Evaluation of Predictions as Individual Probability Density Functions

Resumen de: US20260260180A1

0000 A system and method are disclosed to train machine learning models, generate predictions, and evaluate the predictions as individual probability density functions. Embodiments include a computer comprising a processor and memory and configured to train a first machine learning model to predict a mean demand of one or more items. Embodiments train a second machine learning model to predict a variance associated with the predicted mean demand. Embodiments use the first and second machine learning models and received current sales data to predict a negative binomial variance of demand of the one or more items, comprising a confidence interval specifying a stocking level for the one or more items that will satisfy a defined number of estimated outcomes. Embodiments generate an individual probability density function using the predicted mean demand of one or more items and the predicted negative binomial variance of demand, and evaluate the individual probability density function.

Methods of Operating Electrochemical Storage Devices Based on Anomaly Clustering, and Software and Systems Including Same

Resumen de: US20260259270A1

Methods of operating electrochemical storage devices, such as secondary batteries, battery modules, and battery cells, using machine-learning models for detecting operating conditions that indicate that one or more electrochemical storages device is/are experiencing an anomaly that may affect its operation. In some embodiments such a method may include deploying an anomaly handler that implements a trained clustering model to identify anomalous operating data and using output of the clustering model to take an operation-control action to control an operation of one or more electrochemical storage devices and/or provide an indication that attention may be needed. In some embodiments a trained detector model is deployed to filter out “normal” operating data so that the trained clustering model handles only “anomalous” operating data, which can drive improvements to the anomaly handler. Methods of training machine-learning models and apparatuses and systems implementing anomaly handlers are also disclosed.

PREDICTING PERFORMANCE OF CLINICAL TRIAL FACILITATORS USING PATIENT CLAIMS AND HISTORICAL DATA

Resumen de: US20260260719A1

0000 A clinical trial site evaluation system applies a machine learning technique to predict recruitment performance of a candidate clinical trial facilitator (such as a clinical trial site or a clinical trial investigator) for a clinical trial based on patient claims data or other data associated with the candidate clinical trial facilitator. In a training phase, a training system trains the machine learning model based on historical recruitment data associated with historical clinical trials and patient claims data (or other data) associated with the clinical trial facilitators associated with those trials. In a prediction phase, the machine learning model is applied to claims data (or other data) associated with candidate clinical trial facilitators to predict recruitment performance.

CROSS-DEVICE DATA SYNCHRONIZATION BASED ON SIMULTANEOUS HOTWORD TRIGGERS

Resumen de: US20260260649A1

Techniques are described herein for cross-device data synchronization based on simultaneous hotword triggers. A method includes: executing a first instance of an automated assistant in an inactive state at least in part on a first computing device operated by a user; while in the inactive state, receiving, via one or more microphones of the first computing device, audio data that captures a spoken utterance of the user; processing the audio data using a machine learning model to generate a predicted output that indicates a probability of one or more hotwords being present in the audio data; determining that the predicted output satisfies a threshold that is indicative of the one or more hotwords being present in the audio data; in response to determining that the predicted output satisfies the threshold, performing arbitration with at least one other computing device that is executing at least in part at least one other instance of the automated assistant; and in response to performing arbitration with the at least one other computing device, initiating synchronization of user data or configuration data between the first instance of the automated assistant on the first computing device and the at least one other instance of the automated assistant on the at least one other computing device, the user data comprising data that is based on one or more interactions with the user at the first computing device, the one or more interactions occurring prior to the receiving of the au

DYNAMICALLY DIVIDING ACTIVATIONS AND KERNELS FOR IMPROVING MEMORY EFFICIENCY

Resumen de: US20260260115A1

Embodiments are generally directed to dynamically dividing activations and kernels for improving memory efficiency. An embodiment of a method in a compute engine performing machine learning comprises: receiving, by a convolutional layer of a convolutional neural network (CNN) implemented on the compute engine, a plurality of activation groups contained in an input data, wherein the convolutional layer includes one or more kernel groups and the one or more kernel groups each include a plurality of kernels; determining a plurality of memory efficiency metrics based on the number of activation groups of the plurality of activation groups and the number of kernels of the plurality of kernels; selecting a first optimal number of activation groups and a second optimal number of kernels that are associated with an optimal memory efficiency metric in the plurality of memory efficiency metrics; and performing a convolutional operation on the input data based on the first optimal number and the second optimal number.

MACHINE LEARNING APPARATUS, MACHINE LEARNING METHOD, AND COMPUTER READABLE MEDIUM

Resumen de: US20260260138A1

An inference unit (113) executes inference on input data for each model parameter set using a machine learning model to which the model parameter set is set, and obtains output data indicating an inference result of the machine learning model when the model parameter set has been set. A comparison unit (114) compares features of the inference results between pieces of the output data and obtains a comparison result. An output unit (115) determines information related to information leakage among information included in the inference result indicated in any of the pieces of output data based on the comparison result, applies a modification to the information related to information leakage on the output data, and outputs the modified output data.

Hellinger Decision Trees for Fraud Detection

Resumen de: US20260260174A1

A Hellinger decision tree can detect fraudulent transactions in a data set of financial transactions. Applying the Hellinger decision tree uses a Hellinger distance. The Hellinger decision tree can be part of a machine learning algorithm. In an example, the Hellinger decision tree is a positive and unbalanced Hellinger decision tree used with an imbalanced positive and unlabeled data.

DATA COLLECTION ACROSS MULTIPLE USER EQUIPMENT CAPABILITY TYPES

Resumen de: WO2026180869A1

Example embodiments of the present disclosure are directed to data collection across multiple user equipment (UE) capability types. A method comprises providing, to a second apparatus, results of measurements associated with data collection for a machine learning functionality and ground truth information for the machine learning functionality; receiving, from the second apparatus, further results of the measurements obtained by a third apparatus; and adjusting a measurement performance based on a comparison between the results and the further results.

DATA COLLECTION ACROSS MULTIPLE USER EQUIPMENT CAPABILITY TYPES

Resumen de: WO2026180870A1

Example embodiments of the present disclosure are directed to data collection across multiple user equipment (UE) capability types. A method comprises providing, to a second apparatus, results of measurements associated with data collection for a machine learning functionality and ground truth information for the machine learning functionality; receiving, from the second apparatus, an indication for an adjustment of measurement performance at the first apparatus; and adjusting the measurement performance based on the indication.

INFERENCE DATA SIMILARITY FEEDBACK FOR MACHINE LEARNING MODEL PERFORMANCE MONITORING IN BEAM PREDICTION

Resumen de: US20260260136A1

0000 In an aspect, a UE may obtain a first indication of a first set of characteristics associated with a plurality of reference datasets. The UE may calculate a respective level of similarity between an inference dataset and each of the plurality of reference datasets based on the first set of characteristics and a second set of characteristics associated with the inference dataset. The UE may output at least one second indication of the respective level of similarity between the inference dataset and each of the plurality of reference datasets.

COMPUTER-BASED SYSTEMS CONFIGURED FOR MACHINE-LEARNING CONTEXT-AWARE CURATION AND METHODS OF USE THEREOF

Resumen de: US20260260134A1

0000 Systems and methods of context-aware caller identification via machine learning techniques are disclosed. In one embodiment, an exemplary computer-implemented method may comprise: obtaining a trained activity completion time estimation machine learning model that determines activity completion time prediction data for an activity of an entity; receiving, from a first computing device of a user, current entity-specific device-executed user activity data of a current entity-specific user activity associated with a user and an entity; receiving from a second computing device associated with the particular entity, current user-specific entity activity data associated with a current user-specific entity activity, related to the current entity-specific user activity; utilizing the trained activity completion time prediction machine learning model to determine current user-specific entity activity completion time prediction data for the current user-specific entity activity; and determining a current displaying context to notify the user of the current user-specific entity activity completion time prediction.

SYSTEM AND METHOD FOR MANAGING AN ON-SENSOR MACHINE LEARNING (ML) MODEL

Resumen de: US20260260170A1

0000 A system and method for managing an on-sensor machine learning (ML) model includes monitoring performance parameters of plurality of on-sensor ML models present in an industrial plant; detecting a degradation of at least one on-sensor ML model based on the monitored ML model performance parameters of the plurality of on-sensor ML models, wherein degradation of the at least one on-sensor ML model comprises at least one of: data distribution change, training serving skew, model drift, occurrence of outlier event, and data quality issue; and updating the at least one on-sensor ML model based on the ML model upgradation parameters retrieved from one of the plurality of sources.

METHOD AND APPARATUS FOR SUPPORT OF ARTIFICIAL INTELLIGENCE/MACHINE LEARNING (AI/ML)-BASED POSITIONING

Resumen de: GB2639745A

According to various examples of the present disclosure, there is provided a location management function (LMF) entity configured to: subscribe to or request artificial intelligence/machine learning (AI/ML) -related services from a network data analytics function (NWDAF) entity in relation to an AI/ML model for determining positioning of a user equipment (UE); and receive, from the NWDAF entity, an indication that training has been performed for the AI/ML model. According to various examples of the present disclosure, there is provided a network data analytics function (NWDAF) entity configured to: receive, from a location management function (LMF) entity, a subscription to or request for artificial intelligence/machine learning (AI/ML) -related services in relation to an AI/ML model for determining positioning of a UE; obtain data for training the AI/ML model from at least one other entity; train the AI/ML model based on the obtained data; and transmit, to the LMF entity, an indication that training has been performed for the AI/ML model; wherein the NWDAF entity includes a model training logical function (MTLF).

SYSTEMS AND METHODS FOR GENERATING AND EXECUTING FUNCTION CALLS USING MACHINE LEARNING

Resumen de: US20260252397A1

0000 The methods, systems, and computer networking apparatuses described herein enable language models to receive input (e.g., a query or a request) from a user or application, and without any additional training data or instructions, determine to generate a function call based on the received input from the user and generate the function call based on the determination. In some embodiments, a language model may further access an external tool or application to request an output as a response to the generated function call. The disclosed methods, systems, and networking apparatuses improve the technical field by incorporating language model capabilities within the function calling process, and allowing for function-related information to be provided to a language model via input received in any number of formats or types, including structured or unstructured input, language or non-language input, or any combination thereof.

SCALABLE MACHINE LEARNING PLATFORM WITH INTEGRATED FEATURE GENERATION AND REAL-TIME MODEL SERVING

Resumen de: US20260252904A1

0000 A system and method for machine learning platform management includes a unified architecture for developing and deploying machine learning models at scale. The platform integrates feature generation, model training, and inference services through a centralized interface. The system processes source data through a feature platform to generate training datasets and real-time features. A multi-stage training pipeline enables automated model experimentation through configurable workflows combining core frameworks and user modeling code. The platform implements specialized inference services optimized for high-throughput ranking and recommendation use cases, with distributed feature stores and local caching for efficient feature serving. A comprehensive monitoring system tracks model performance, feature distributions, and prediction quality through automated anomaly detection. The platform enables rapid experimentation while maintaining production reliability through automated deployment orchestration, optimized inference engines, and continuous feedback loops for model improvement.

Transformer-Based Assistant for Identifying, Organizing, and Responding to Customer Concerns

Nº publicación: US20260252607A1 27/08/2026

Solicitante:

UJWAL INC [US]
Ujwal Inc.

Resumen de: US20260252607A1

0000 Transformer-based agent assistant systems as machine learning-based customer service tools that analyze past customer-agent conversations to build a knowledge base of problem-resolution steps are disclosed. The system may include a natural language processing (NLP) model and a transformer-based model to extract and generate customer concerns and resolutions. One embodiment also includes a head-topic and subtopic detection module for identifying trends in customer concerns. Another embodiment uses a question-answering model and a zero-shot-NLI (natural language inference) classifier for entity extraction and detection. The system is designed to be flexible, incorporating new data over time, and can retrieve company documentation or FAQs for the agent based on cosine similarity.