Alerta

ARTIFICIAL INTELLIGENCE BASED METHODS AND SYSTEMS FOR ANOMALY PATTERN DETECTION IN DATASETS

Resumen de: WO2026212789A1

A process includes obtaining a dataset having a plurality of data records that include time series data, determining a typology feature for a typology of interest of the dataset based on an exponentially decaying velocity of at least a portion of the plurality of data records. In response to the typology feature satisfying a typology condition, the process may provide input data features associated with the dataset that include at least a data feature relating to the exponentially decaying velocity of the portion of the plurality of data records into a machine learning anomaly classification model. The machine learning anomaly classification model may be operated on the input data features. In response to an output of the machine learning anomaly classification model satisfying an anomaly condition, an anomaly indicator that indicates that the dataset is an anomaly is stored in storage.

AUTOMATED FEATURE GENERATION FOR PREDICTIVE ANALYTICS WITH GENERATIVE ARTIFICIAL INTELLIGENCE

Resumen de: US20260310654A1

A system for automated generation of features for a machine learning model obtains a machine-readable schema of a training dataset and generates candidate feature concepts by providing at least one first prompt, including at least one seed value not derived from the training dataset, to a first large language model operated at a first randomness setting. The system generates normalized feature specifications using a second large language model operated at a second randomness setting lower than the first randomness setting, and generates feature computation code using a third large language model fine-tuned on feature computation code. Prior to storage or execution, the system performs schema-aware validation by parsing the feature computation code into an abstract syntax tree, extracting identifier references, enforcing an identifier allowlist, verifying type rules, and rejecting prohibited constructs. Validated code is executed in a sandboxed execution environment enforcing runtime restrictions.

PARALLEL PROCESSING FOR MODEL EXECUTION

Resumen de: US20260311057A1

The subject technology provides for distributed training and inferencing of a machine learning model across multiple processing resources. During a forward pass, each processing resource generates a portion of an update for its corresponding portion of a feature set. While performing this computation, each resource can approximate, compress, and transmit its portion of the features to other resources, and can receive an approximated and compressed portion generated elsewhere. Each resource then decompresses the received representation and updates its local feature portion without waiting for full-precision data from other nodes. By allowing approximation, compression, communication, and computation to occur in parallel, the system may reduce communication overhead that typically causes idle compute periods. This structure may improve sustained utilization and responsiveness when models operate across multiple distributed compute nodes.

GENERATING SPEECH TEST MATERIALS WITH CUSTOMIZABLE LINGUISTIC CHARACTERISTICS

Resumen de: WO2026210286A1

Presented herein are systems and methods for generating one or more test sentences. A labeled training set including one or more test sentences with corresponding labels is obtained. A machine learning model is trained based on the labeled training set to generate a pre-trained machine learning model. The pre-trained machine learning model is fine-tuned, based on a fine-tuning sentence set, to generate a fine-tuned machine learning model. A user prompt indicating a request for one or more test sentences exhibiting one or more linguistic characteristics is obtained. The fine-tuned machine learning model is configured to generate one or more test sentences based on the request indicated in the user prompt.

MACHINE LEARNING TECHNIQUES FOR DISCOVERING KEYS IN RELATIONAL DATASETS

Resumen de: US20260310634A1

0000 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).

OPTIMIZING ARTIFICIAL INTELLIGENCE (AI) MODEL TRAINING AND INFERENCING PERFORMANCE THROUGH SERVICE CLASS PRIORITIZATION

Resumen de: US20260311043A1

0000 In an embodiment, optimizing AI model training and inferencing performance through service class prioritization includes detecting one or more patterns in a plurality of service classes previously executed on a computing system, wherein each service class in the plurality of service classes includes a respective set of one or more work items; determining, based on the one or more patterns, a pattern strength score for each service class in the plurality of service classes; selecting, based on the pattern strength score, a service class for inclusion in a training data set; and training a machine learning model using the training data set.

COMPUTER SYSTEM AND METHOD FOR DETERMINING A CAUSE OF AN ANOMALY EVENT FOR A NETWORK DEVICE

Resumen de: US20260312962A1

A computing system and process for training a machine-learning (ML) model for determining a cause of a detected anomaly event for at least one network device. Accessed are training datasets including a plurality of anomaly events for a plurality of network devices and a plurality of computer network events associated with one or more computer networks. A determination is made regarding an appearance frequency a certain anomaly event for a network device occurs within a defined time period a network event occurs. A determination is made regarding a probability a certain anomaly event for a network device is associated with a network event responsive to the determined appearance frequency, such that, when the ML model is executed, a determination is made for a likelihood a given anomaly event for a network device is caused by a given network event.

ULTRA-HIGH-THROUGHPUT MOLECULAR RETRIEVAL SYSTEM FOR VIRTUAL SCREENING OF LARGE CHEMICAL LIBRARIES

Resumen de: WO2026213359A1

A system and method may include a machine learning model configured to generate vector embeddings of molecules based on molecular representations, a storage system configured to store embeddings corresponding to a plurality of molecules in the chemical library, and a similarity search engine configured to compute similarity between a query molecule embedding and stored embeddings. A system may include a computing infrastructure configured to perform the similarity computation across at least one distributed computing node, wherein the system is configured to identify molecules having similarity above a threshold relative to the query molecule.

SYSTEM AND METHOD FOR ANALYZING AND NORMALIZING SLEEP DATA

Resumen de: WO2026213356A1

A system and method for standardizing and analyzing sleep data from a plurality of heterogeneous third-party device sources are disclosed. The system ingests sleep data from wearable devices, mobile applications, and aggregated health data platforms under a Bring Your Own Device (BYOD) framework. A machine learning-based standardization module normalizes heterogeneous sleep data into a unified dataset by compensating for differences in data formats, sampling rates, and device-specific collection methodologies, and predicts missing values to ensure dataset completeness. A composite predictive model comprising an ensemble of machine learning algorithms cross-validates sleep parameters, applies oversampling to address imbalanced training data, and generates a sleep score representing the user's sleep quality. A feedback mechanism receives user-reported sleep quality input and incorporates it into continuous refinement of the standardization module. Personalized model adjustments using parallel multi-device recordings further improve predictive accuracy for individual users.

REMOTE VALIDATION OF AUTOMATED TELLER MACHINES (ATMs) USING MACHINE LEARNING

Resumen de: US20260311570A1

0000 Aspects of the disclosure relate to a remote validation platform for ATMs. The platform may train a reconfiguration model to output sets of configuration instructions. The platform may receive a first information stream from a first ATM. The platform may receive one or more additional information streams from sources associated with the first ATM. The platform may generate, using the reconfiguration model, a first set of configuration instructions based on the first information stream and the one or more additional information streams. The platform may cause reconfiguration of the first ATM based on the first set of configuration instructions. The platform may refine the reconfiguration model based on the first set of configuration instructions.

ML-BASED REFINEMENT MODEL FOR DEVELOPMENT OF A WELLNESS ASSESSMENT METHOD

Resumen de: WO2026210466A1

The present disclosure relates to a machine learning (ML) based refinement model for development of a wellness assessment method. More particularly, the model acquires genetic, metabolic, anthropometry, clinical data and live tracking data from health devices to provide personalized diet and exercise charts for subjects. Furthermore, the method is a tier-based process wherein tier 1 employs LASSO for regression analysis; tier-2 involves ML based decision models for data validation and tier-3 incorporates CNN and LLM to decide the subtype a subject belongs to. The proposed Machine Learning (ML) based model incorporates Natural Language Processing (NLP) and Large Language Models (LLM) to assess behavioral and recorded language data and links the same to the subtypes generated using genetic, metabolomic and anthropometry data.

AI Superintelligence Application State Machine

Resumen de: US20260311258A1

0000 The present invention provides a method and apparatus for a thinking system using an ASI distributed Application State Machine that connects to real, accurate, dedicated absolute data, providing privacy, security, and eliminating hallucinated AI. Specifically, one embodiment of the present invention discloses an ASI Command, Control, Communications and Computer Intelligence (C4I) system for independent, intelligent AI Prompting and ASI Browsing. The ASI Application State Machine comprises: ASI Application data structure network; ASI Enterprise Interface State Machine, said data structures interfacing with a back-end channel, a Finite State Machine embedded in each data structure; and Application Service Information Base providing a uniform interface for ASI Application data structure identities. Additional embodiments disclose ASI Browser displaying ASI Applications; ASI Agents; ASI Application Network Nodes; ASI Operating System; ASI Switch; ASI Database of real, accurate, dedicated absolute data, for Machine Learning to train AI Models, for performing real-time, bi-directional transactions, connecting to billing.

ONBOARD FAULT DIAGNOSIS AND RECOVERY IN WIRELESS SENSOR NODE DEVICE AND METHOD

Resumen de: WO2026212573A1

Wireless sensor node devices and methods in wireless sensor node devices are disclosed A wireless sensor node device comprises: a sensor configured to generate sensor data; a wireless network interface configured to transmit the sensor data from the sensor node device; and a controller comprising: a machine learning classifier configured to identify anomalous data in the sensor data and to classify the anomalous data; a fault diagnosis module configured to diagnose a fault source based on a classification of the anomalous data; and a recovery module configured to generate a recovery action indication based on the fault source.

DEVICE TELEMETRY PROCESSING SYSTEM WITH PROBABILISTIC CLASSIFICATION AND PREDICTIVE HEALTH MODELING

Resumen de: US20260310574A1

0000 Systems and methods for managing mobile device lifecycle data by collecting disparate data from external downstream sources and combining it with deep device-level data captured directly from individual devices including telemetry, hardware diagnostics, and physical access data. The combined data trains multiple machine learning models applied to individual and aggregate devices throughout the device lifecycle. A time-series analysis model processes device telemetry data to generate a cumulative stress score representing aggregated device wear, and a survival analysis model generates an estimated remaining life prediction, both stored in a device passport data structure linked to a device identifier. A probabilistic classifier with uncertainty scoring routes devices through bypass or standard processing pathways based on fault prediction confidence. A cross-modal reasoning engine employing multimodal large language models correlates diagnostic logs with device photographs to generate unified device condition assessments.

REAL-TIME TELEMETRY SHARING BETWEEN QUANTUM MANAGEMENT INSTANCES

Resumen de: US20260311020A1

0000 One example method is directed to quantum job orchestration, and includes obtaining, by an orchestration instance, telemetry from respective control planes associated with one or more QPUs (quantum processing units), providing the telemetry as input to an ML (machine learning) model of the orchestration instance, using the ML model to perform an inferencing process whose output is a profile that comprises information usable to make an orchestration decision for a job, based on the profile, orchestrating the job to a particular one of the QPUs, and broadcasting the profile to one or more other orchestration instances.

Centralized Multi-Source Incident Data Clearinghouse Architecture for Incident Intelligence and Cross-System Event Correlation

Resumen de: US20260310681A1

A centralized multi-source incident data clearinghouse receives heterogeneous incident data from public safety and non-emergency service systems, normalizes and correlates the data into unified incident profiles independent of any individual source, distributes enriched structured outputs via a shared integration layer that abstracts direct point-to-point integrations, reduces duplicated computation and latency, and incorporates operational outcome feedback into adaptive machine-learning refinement without performing call routing.

PROCESS AND SYSTEM FOR GENERATION OF COKE DRUM OUTAGE PROJECTIONS

Resumen de: US20260310246A1

0000 An apparatus includes at least one processing device configured to obtain sensor data from sensors associated with one or more coke drums in a refinery. The at least one processing device is also configured to generate an outage projection for an end of a feed cycle of at least one of the coke drums by processing, utilizing a machine learning model, information derived from the sensor data characterizing (i) an amount of feedstock which has been fed to the at least one coke drum during the feed cycle, (ii) a time at which a level indicator for the at least one coke drum first exceeds a designated threshold value, and (iii) a density of the feedstock fed to the at least one coke drum during the feed cycle. The at least one processing device is further configured to provide, at a user interface, the generated outage projection.

COMPLETE HUMAN-IN-THE-LOOP AGENTIC AI SYSTEM FOR CHURN PREDICTION AND IN-TIME RETENTION PLANNING AND EXECUTION

Resumen de: US20260310944A1

A computer-implemented method for customer retention uses an agentic artificial intelligence pipeline comprising an orchestration layer that coordinates a plurality of specialized AI agents via a schema-validated inter-agent communication protocol. Customer data may be collected and processed through the AI agents. A custom churn modeling agent may automatically select and train a machine learning algorithm and generate local and global explanations for each churn prediction. Survival analysis techniques, including Kaplan-Meier estimators, may estimate when each at-risk customer is likely to churn. A Bayesian Network may be constructed to identify probable churn causes and map them to corresponding retention treatments. A Large Language Model agent conditioned on the identified causes and treatments may generate personalized retention communications. Retention recommendations may be presented to a human operator for approval prior to implementation. Observed retention outcomes may be used to update Bayesian Network conditional probability distributions to improve future recommendations.

ERRONEOUS ANSWER RISK DETERMINATION DEVICE, ERRONEOUS ANSWER RISK DETERMINATION METHOD, AND PROGRAM

Resumen de: WO2026210775A1

Provided is an erroneous answer risk determination device capable of accurately determining an erroneous answer risk of a machine learning model. The erroneous answer risk determination device includes an erroneous answer risk determination unit that determines an erroneous answer risk on the basis of a result of comparing a first set and a second set, the first set being a set of errors in an estimation function, which is an estimation result of an unknown function, allowed on the basis of correctness of an output domain with respect to M input domains in a first composite function obtained by composing the estimation function and a first known function, and the second set being a set of errors in the estimation function allowed on the basis of correctness of an output domain with respect to M input domains in a second composite function obtained by composing the estimation function and a second known function different from the first known function.

DQN-BASED TASK PLANNING

Resumen de: US20260310507A1

0000 One example method includes receiving a mission in natural language format, which specifies an action to be executed by an agent on behalf of a user. A first embedding operation is performed on the mission to create a first vector representation. From a pool of tasks, a first task is selected for the agent to execute the mission. A second embedding operation generates a second vector representation of this task, while a third embedding operation produces a third vector representation of the task's execution order. These vector representations are concatenated to form a single vector. A machine learning model then utilizes this concatenated vector to generate a task plan. This plan details the first task and its execution order, along with additional tasks and their respective execution orders, guiding the agent in mission completion.

SYSTEMS AND METHODS FOR FEDERATED LEARNING ENABLED 5G/6G MULTI-TASK SEMANTIC COMMUNICATIONS FRAMEWORK

Resumen de: US20260312879A1

Example implementations encompass systems and methods for federated learning-assisted multi-task semantic communications tailored for 5G/6G wireless systems. A Centralized Server (CS) hosts a knowledge base with interconnected Machine Learning (ML)-based semantic codec models, extracting and synthesizing semantic information across various tasks using single or multi-modal observations. The CS connects with Decentralized Servers (DSs) and Edge Servers (ESs), which handle task-related data transmission and reception. This setup enables training of global models for semantic communications through federated learning. ML models in DS knowledge bases allow local model updates, leveraging ES-assisted transmissions via a Local Model Aggregator (LMA). The CS aggregates model weights from the DSs using a Global Model Aggregator (GMA). Semantic transmission requests can be initiated from ESs and forwarded to the CS, which responds by providing semantic encoding and decoding models to ensure efficient communication of semantic information between involved parties.

METHOD, MEDIUM, AND SYSTEM FOR VIRTUAL AGENTS TO HELP CUSTOMERS AND BUSINESSES

Resumen de: US20260311229A1

A system and method are provided for dynamically selecting and invoking application programming interfaces (APIs), tools, functions, and workflows based on natural-language inputs and contextual user information. A virtual agent server maintains a semantic registry comprising identifiers, parameter definitions, and semantic annotations associated with a plurality of APIs and executable actions. A natural-language input is processed to determine a desired action and contextual information. Runtime semantic matching is performed between the natural-language input and semantic annotations to dynamically select one or more executable actions without use of a precompiled intent-specific routing definition. Parameter values are derived using dialog histories, contextual inference, runtime context information, and machine learning models. The system transmits a invocation request configured to cause a client application or external execution system to invoke the selected API, tool, function, or workflow, or alternatively executes the selected executable action on a server-side execution system using the derived parameter values.

COMPUTING SYSTEM AND METHOD FOR CREATING A MACHINE LEARNING MODEL HAVING IMPROVED FAIRNESS

Resumen de: US20260311045A1

A computing platform installed with executable software for producing a fairer version of a machine learning model comprising a perturbed version of additive explanations for the model's features in which the additive explanations are perturbed by corresponding perturbation coefficients, where the computing platform is programmed to (i) produce candidates for the perturbation coefficients by carrying out a training process for a trainable object comprising the perturbation coefficients as trainable parameters, where the training process uses an objective function having an error term, a bias term, and a bias coefficient and involves multiple training rounds that are carried out using different bias-coefficient values and produce different candidates for the perturbation coefficients, (ii) evaluate the performance and bias achieved by the different candidates for the perturbation coefficients, and (iii) based on the evaluation, determine a given candidate for the coefficient values to use in the fairer version of the machine learning model.

COORDINATING INTERACTIONS WITH HETEROGENOUS MACHINE LEARNING AGENTS

Resumen de: WO2026212706A1

This document relates to facilitating interactions between a user and multiple agents, such as agents that provide different generative machine learning capabilities. In certain disclosed implementations, user input is received and two or more agents are selected to perform a particular task for the user. Then, interactions with the agents are coordinated on behalf of the user based on agent definitions of the selected agents. A result of the particular task is received and provided to the user. For instance, the user input can be received via a chat interface that allows the user to conduct a single chat session that involves the particular task being performed by multiple agents, thus giving the user the appearance of interacting with a single agent.

SYSTEMS AND METHODS FOR ITERATIVE MACHINE LEARNING MODEL OPTIMIZATION THROUGH ADAPTIVE PARAMETER AD JUSTMENT

Nº publicación: WO2026212917A1 08/10/2026

Solicitante:

AML JV LLC [US]
AML JV, LLC

Resumen de: WO2026212917A1

Described herein are systems and methods that related to techniques for iterative machine learning model optimization through adaptive parameter adjustment. In an example, a system can be configured to obtain input data that includes at least one user input specifying a type of endpoint. Based on the input data, the system can identify features from a set of features to use in evaluating multiple models, with each feature corresponding to diagnostic criteria used to represent aspects of treatment profiles maintained in a database. The system can train and evaluate performance of each of the models and generate a user interface to display performance metrics for the models. In some examples, this user interface can be provided to a device that generated the input data to show the performance metrics for the models.