Alerta

ARTIFICIAL INTELLIGENCE-BASED NETWORK PERFORMANCE PREDICTION TRAINING

Resumen de: US20260281753A1

0000 Certain aspects of the present disclosure provide techniques for training an artificial intelligence-based wireless communication network simulator. An example method includes obtaining first training data associated with a first network simulator, wherein the first training data includes first parameters associated with simulation of a wireless network operating, in a scenario, according to first configurations, wherein the wireless network includes a network node in communication with a user equipment (UE); obtaining second training data associated with a second network simulator, wherein the second training data includes second parameters associated with simulation of the wireless network operating, in the scenario, according to second configurations; generating third training data based at least in part on the first training data and the second training data; and training a machine learning model, associated with prediction of performance metrics of the wireless network, based at least in part on the third training data.

REAL-TIME MEDICAL CODE VALIDATION USING CONTRASTIVE LEARNING AND RETRIEVAL-AUGMENTED GENERATION

Resumen de: US20260279529A1

0000 A system validates medical claim coding information during a coding workflow and before claim submission. The system receives diagnosis and procedure codes for a healthcare encounter and obtains embeddings produced by a contrastive machine-learning model trained on labeled valid and invalid code combinations. An embedding-based compatibility measure is determined using a similarity computation, and multi-code weighting generates an aggregated compatibility determination for encounters involving multiple codes. The aggregated determination is compared to a threshold to assign a preliminary classification of valid, invalid, or needs review. Policy or guidance passages are retrieved and used to confirm, revise, or override the preliminary classification, producing a final classification. A policy-cited explanation is generated from the retrieved passages. The system outputs the final classification and explanation to a user interface, enabling correction of potential coding errors before claim submission.

MODIFIED MECHANICAL SPECIFIC ENERGY SYSTEM AND METHODS USING A TERNARY GRAPH-BASED APPROACH

Resumen de: US20260278331A1

0000 The Modified Mechanical Specific Energy (MSE 3.0) system introduces an advanced, AI-driven drilling optimization approach. Unlike traditional models, MSE 3.0 integrates a ternary graph-based visualization to dynamically analyze cuttings morphology, pipe vibration, and MSE in real time. The system utilizes machine learning algorithms for predictive analytics, second-order optimization for early inefficiency detection, and a cost-effective surface-based measurement approach. By eliminating reliance on downhole sensors, MSE 3.0 reduces operational costs while enhancing drilling efficiency, making it applicable to hydrocarbon, geothermal, and critical mineral extraction operations. A key innovation of MSE 3.0 is the integration of real-time Cuttings Morphology Analysis, providing a direct, continuous assessment of downhole conditions and formation response. This enables precise, data-driven optimization of drilling patterns, significantly improving predictive accuracy and wellbore stability over conventional single-variable MSE models.

Synthesis of Air Quality Data from Diverse Sources Using Machine Learning

Resumen de: US20260276614A1

0000 Composite health impact scoring of air quality for multiple pollutants is described. In one or more implementations, input data is received indicating concentrations of multiple pollutants at a geographic location. A composite health impact score is generated based on the input data. As part of this, response factors are applied to the multiple pollutants, where the response factors quantify health impacts from exposure to respective pollutants of the multiple pollutants. Moreover, independence factors are applied to the multiple pollutants, where the independence factors adjust the health impacts attributable to the respective pollutants based on combined exposure to the multiple pollutants. The composite health impact score is communicated to a client device for output in a user interface.

METHOD AND SYSTEM FOR DYNAMICALLY CUSTOMIZING FOOD FORMULATIONS

Resumen de: US20260278461A1

The present disclosure provides a system, a method and a non-transitory computer-readable storage medium for dynamically customizing one or more pre-existing food formulations. The method includes receiving one or more input parameters from a user device. The method includes extracting one or more pre-existing food formulations and a set of food processing conditions for the one or more pre-existing food formulations. The method includes generating, using the machine learning model, the nutrient degradation profile for each nutrient corresponding to one or more ingredients associated with the one or more pre-existing food formulations. The method includes determining a set of degradation coefficients for each nutrient corresponding to the one or more ingredients. The method includes predicting nutrient retention rates associated with the one or more pre-existing food formulations. The method includes customizing the one or more pre-existing food formulations.

ENTITY MATCHING BY LARGE LANGUAGE MODELS WITH RUN TIME ANNOTATION

Resumen de: US20260277882A1

A computing device may select, at run time, a number of pairs of records from a plurality of records. Each record, of the plurality of records, includes attributes. The computing device may obtain, for each of the selected number of pairs of records, key attributes at run time. The computing device may provide, at run time, a prompt to a machine learning model to obtain a prediction of whether records, of a particular pair of records that is not selected from the records, are matched. The prompt may be generated based on the key attributes of the selected number of pairs of records.

IMPEDANCE-BASED CLASSIFICATION OF BIOLOGICAL CELLS OR OTHER PARTICLES VIA MACHINE LEARNING METHODS

Resumen de: US20260279571A1

The disclosure includes a platform for classifying particles (e.g., biological cells) based on impedance measurements of individual particles using a biodevice. A machine learning model can be used to classify cells with high accuracy (e.g., greater than 95%). This can be accomplished without the need for sample preparation or cellular tagging. The biodevice can immobilize individual cells (e.g., mechanically immobilize with a micro-pore). An external AC waveform can be applied, for example, in the 1 Hz to 1 MHz range. The machine learning model may be generated to provide classifications based on, for example, measurements of impedance magnitude and phase angle at discrete frequencies.

COMPUTER-BASED SYSTEMS CONFIGURED TO ORCHESTRATE A TRANSFER OF A DATA OBJECT AND METHODS OF USE THEREOF

Resumen de: WO2026193487A1

In some embodiments, the present disclosure provides an exemplary method that may include steps of receiving a request for data communication from an external entity to a particular internal enterprise system; identifying the distinct internal system-specific data ontology associated with the particular internal enterprise system; utilizing a machine learning matrix to dynamically compare the request for the data communication to the distinct internal system-specific data ontology; generating a standardized data ontology; verifying the plurality of protocols associated with the standardized data ontology; generating a standardized application programming interface between the external entity and the particular internal enterprise system; and utilizing an adapter to orchestrate a transfer of a data object from the particular internal enterprise system to the external entity.

NOISY QUANTUM CHANNEL ERROR MITIGATION

Resumen de: US20260278451A1

0000 Systems and methods for implementing noisy quantum channel error mitigation are described. For example, a method can include providing to a machine learning model, input data indicative of an original quantum state of a quantum system that is modified by an amount of noise added by a non-unitary quantum communication channel. The machine learning model is trained to predict the original quantum state in order to perform error mitigation by simulating an inverse of the non-unitary quantum communication channel to reverse one or more errors caused by the non-unitary quantum communication channel. The method can further include obtaining, from the machine learning model, output data indicative of the original quantum state, and causing a quantum computing system to perform at least one quantum computing operation, involving at least one of error mitigation or quantum state preparation, based on the output data.

RESIDUAL INITIALIZATION FOR TRAINING NODE NETWORKS

Resumen de: US20260278384A1

0000 This disclosure describes systems, methods, and computer-readable media for initializing and training differentiable logic gate networks for machine-learning inference. In various embodiments, parameters of learnable logic gate operators are initialized using a residual initialization that biases the initial probability distribution of discrete gate choices toward a feedforward operation, such as a wire or an inverter, thereby preserving information flow and mitigating vanishing gradients in deep networks. During training, the gate parameters are updated so that gates can remain as feedforward connections or transition to other logic operations as needed. The residual initialization can be applied to convolutional logic gate-tree architectures and pooling operations, and trained networks can be discretized and synthesized into efficient logic circuits in hardware or software.

SYSTEMS AND METHODS FOR PREDICTING DRY MATTER INTAKE IN ANIMALS USING MACHINE LEARNING

Resumen de: US20260271888A1

A computing system may receive data comprising values of a set of variables corresponding to a plurality of animals, the set of the variables including a water intake and a dry matter intake. A computing system can train a machine learning model on the data to provide values of dry matter intake based on the values of the set of the variables. A computing system can after training, provide the machine learning model with input values of at least a subset of the variables associated with at least one animal to generate a value of dry matter intake for the at least one animal.

METHOD FOR DETECTING PRESENCE OF AN ELECTRIC ARC AND ASSOCIATED DEVICE

Resumen de: US20260276691A1

A computer-implemented method for detecting the presence of an electric arc on a power line using a machine learning algorithm, carried out using a time-correlated measurement of the arc voltage Varc, of the current I and of the source voltage Vsource and also a knowledge model for linking the arc voltage Varc and the current I to constants depending on the knowledge model, implemented after a learning phase that involves adjusting a learning model that takes the current I and the source voltage Vsource at input and provides the voltage Varc and the constants of the knowledge model at output; the method including continuously detecting the presence of an arc using a decision function depending on a cost function used in the learning phase.

Schema Transformation For Managing An Application Build

Resumen de: US20260277602A1

Techniques for generating a schema transformation for application data to monitor and manage the application in a runtime environment are disclosed. A system runs an application plugin in a runtime environment to identify data generated by application modules in one or both of an application build process and an application execution process. The application plugin is a software program executed together with the application build process. The application plugin identifies a source schema associated with application data. The application plugin identifies a target schema associated with an analysis program or machine learning model. The application plugin generates a schema transformation to convert application runtime data into a target data set. The system applies the target data set to an analysis program, such as a machine learning model, to generate output analysis data associated with the application.

Mitigating cyber-threats in a network

Resumen de: GB2704692A

A method for mitigating cyber-threats in a target network comprising a plurality of nodes. A decentralised multi-agent model with a plurality of local models is deployed to the target network S210. Each of the plurality of local models monitors a local region of the target network to obtain local network information S220 and predicts threat mitigation actions based on the local network information S230. The decentralised multi-agent model may be a machine learning model, trained using reinforcement learning. The local models may be trained using one or more training networks, using a trainer system which iteratively evaluates a performance of the multi-agent model and adjusts the local models. Figure 2

Machine learning with PII protection and explainability

Resumen de: GB2704709A

Processing input data 300, comprising receiving an encrypted model parameter update, wherein at least model parameters relating to personally identifiable information (PII) are encrypted using a homomorphic encryption algorithm 302. The encrypted model parameter update is decrypted using the homomorphic encryption algorithm 304. The decrypted model parameter update is applied to model parameters of a machine learning model 306. The machine learning model receives input data comprising personally identifiable information 308. A machine learning model is used to process the input data to produce output data 310. Finally, an explainability algorithm is applied to the output data 312. The receiving of the encrypted model parameter update may comprise the receiving of an encrypted model parameter update from a server device that aggregate the encrypted model parameter update from local encrypted model parameter updates received from a plurality of user devices. Fig. 3

A METHOD FOR GENERATING TEST CASES FOR TESTING A SOFTWARE PROGRAM

Resumen de: EP4807565A1

0001 The invention relates to a method (100) for generating test cases for testing a software program, comprising: - Training (101) a first machine learning model (1) for a next-token prediction based on a training data set, the training data set comprising test cases for testing the software program in a code format, - Providing (102) a labelled training data set, the labelled training data set comprising test cases that are each associated with at least one characteristic of the software program, - Performing (103) a respective fine-tuning of the trained first machine learning model (1) for the next-token prediction based on the labelled training data set for each characteristic of the software program to provide a specialized machine learning model (2) for each respective characteristic, - Generating (104) the test cases for testing the software program for at least one of the characteristics of the software program by utilizing the first machine learning model (1) and/or a respective specialized machine learning model (2) that was fine-tuned for said at least one of the characteristics.

MACHINE LEARNING BASED SEMICONDUCTOR MEASUREMENT MODELS TRAINED USING HISTORICAL DATA

Resumen de: US20250238690A1

Methods and systems for using historical measurement data to train a present state, machine learning (ML) based measurement model are described herein. This approach takes advantage of the correlation between structural characteristics of measured samples fabricated in accordance with different design revisions, process revisions, or both. In one aspect, a present state, ML based measurement model is trained using training data associated with measurements of a plurality of instances of a current version of a semiconductor structure in a present state of a semiconductor process flow and training data associated with measurements of a plurality of instances of a historical version of the semiconductor structure in the present state of the semiconductor process flow. In some examples, training data also includes prior state measurement data. Historical training data, prior state training data, or both, may be derived from actual reference measurements, in-line, production measurements, or both.

FLEXIBLE MACHINE LEARNING MODEL COMPRESSION

Resumen de: WO2025101721A1

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for compresses a machine learning model having a plurality of parameters. In one aspect, one of the methods includes obtaining trained values of a set of parameters for at least a portion of a machine learning model; identifying one or more dense ranges for the trained values; determining a least number of bits required to represent each trained value within the one or more dense ranges; identifying a second format having a range that is smaller than a range of the first format; and generating a compressed version of the at least a portion of the machine learning model.

Control parameter optimization

Nº publicación: GB2704660A 16/09/2026

Solicitante:

MOTORWAY ONLINE LTD [GB]
Motorway Online Ltd

Resumen de: GB2704660A

A computer system comprises a parameter optimization module 102 for optimizing input parameters of a process having a binary output, the module comprising: probabilistic machine learning model 1021 trained to model a relationship between a first and a second controllable input parameter of the process and the binary output, the relationship defining an optimized pair of input parameter values 104; and batched binary Bayesian Optimizer 1022 configured to generate, from the trained probabilistic machine-learning model, a list 105 comprising training pairs of input parameter values. The parameter-optimization-module 102 is configured to output the optimized pair of input parameter values 104 and the list 105 of training pairs of input parameter values, and to receive binary output values of the process performed using the training pairs of input parameter values. The computer system is configured to periodically retrain to generate a new optimized pair 104 of input values and a new list 105 of training pairs of input parameters. Also claimed is assigning received requests to a first or second group to perform the process using the optimized or training input parameters, respectively. Utilizing batched binary Bayesian optimization creates a much smaller training data set by selecting training parameters with the greatest impact when model 1021 is retrained. The computer system operates using optimized input parameters 104 that produce desired results, yet still producing new trai