Alerta

METHOD AND SYSTEM FOR HYPERGRAPH NEURAL NETWORK INFERENCE

Absstract of: US20260278332A1

A method and system for hypergraph neural network inference are provided, wherein the method includes: extracting topological information of vertices with the highest degree from a compressed hypergraph topological representation, identifying common high-degree vertices and their connected hyperedges, and transmitting them to a task queue; based on common vertices from the task queue and a common vertex table, triggering hyperedge aggregation operations on each hyperedge in parallel; temporarily storing intermediate aggregation results in an intermediate result buffer of an on-chip cache unit after the aggregation, and reusing the intermediate results of the common vertices during sequential execution based on a hyperedge storage order, thereby obtaining a final aggregation result for each hyperedge. This approach enables compact execution of computing tasks and improves inference efficiency by reducing unnecessary memory access and communication through cache reuse.

DEEP HARMONIC FINESSE: SIGNAL SEPARATION IN WEARABLE SYSTEMS WITH LIMITED DATA

Absstract of: US20260278330A1

A method for separation of non-stationary quasi-periodic signals when limited data is available is described. The method utilizes prior knowledge of time-frequency patterns in the signals to mask and in-paint spectrograms. In one implementation this is achieved through an application-inspired deep harmonic neural network coupled with an integrated pattern alignment component. The network's structure embeds the implicit harmonic priors within the time-frequency domain, while the pattern-alignment method transforms the sensed signal, ensuring a strong alignment with the network.

SYSTEM, METHOD, AND APPARATUS FOR RECURRENT NEURAL NETWORKS

Absstract of: US20260277546A1

A method for computation with recurrent neural networks includes receiving an input drive and a recurrent drive, producing at least one modulatory response; computing at least one output response, each output response including a sum of: (1) the input drive multiplied by a function of at least one of the at least one modulatory response, each input drive including a function of at least one input, and (2) the recurrent drive multiplied by a function of at least one of the at least one modulatory response, each recurrent drive including a function of the at least one output response, each modulatory response including a function of at least one of (i) the at least one input, (ii) the at least one output response, or (iii) at least one first offset, and computing a readout of the at least one output response.

NEURAL NETWORK MODEL TRAINING METHOD BASED ON PERIODIC FEATURE AND NEURAL NETWORK MODEL

Absstract of: WO2026189008A1

The present application provides a neural network model training method based on a periodic feature and a neural network model. The neural network model comprises an input layer and a self-attention layer. The neural network model training method based on a periodic feature comprises: calling an input layer to extract a plurality of pieces of first feature data of a training sample; performing linear transformation on the plurality of pieces of first feature data to obtain a periodic feature and an aperiodic feature; decomposing the periodic feature into a sine wave feature and a cosine wave feature; concatenating the aperiodic feature, the sine wave feature, and the cosine wave feature to obtain a concatenation result corresponding to the training sample; and using the concatenation result as an input of a self-attention layer to train a neural network model to be trained to obtain a target neural network model, the target neural network model having a function of extracting a periodic feature of data to be processed. In the embodiments of the present application, out-of-distribution periodic features of training samples can be effectively captured, thereby greatly improving model training efficiency.

METHOD AND DEVICE FOR PLAYING MUSIC USING BLOCKS

Absstract of: US20260279315A1

Disclosed are a method and a device for playing music by using blocks. The music playing method using at least one blocks, which is performed by the music playing device according to an embodiment of the present disclosure, may include capturing an image including the at least one or more blocks through a reception device mounted on the music playing device, recognizing the captured at least one or more blocks by using a trained deep learning neural network model, determining an arrangement structure of the recognized blocks, and playing the music at the music playing device, based on the determined arrangement structure.

CONTRASTIVE IMAGE CAPTIONING NEURAL NETWORKS WITH VECTOR QUANTIZATION

Absstract of: WO2025128886A1

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training an image processing neural network to generate discrete representations of input images. The discrete representations can then be used for any of a variety of downstream tasks.

METHOD FOR DETECTING INFORMATION ABOUT OBJECTS AND OBJECT PARTS IN A SPACE

Absstract of: EP4807699A1

Method for detecting information about objects and object parts in a space using a LiDAR sensor, which has a field of view and in which measurement data is converted into a 3D model of information about objects in the field of view, comprising the steps:a.) acquiring a 3D point cloud by the LiDAR sensor in the field of view,b.) converting the 3D point cloud to a 2D digital image in a calculating component,c.) providing the 2D digital image to a detecting component which comprises a trained neural network module,c.1) whereas the module estimates basic geometrical 3D lines of objects, preferably including endpoints of lines andc.2) whereas the module establishes a model of information about objects in the field of view based on the model of 3D lines.

FINGER PRESSING STATE DETECTION METHOD, AND TRAINING METHOD, ELECTRONIC DEVICE AND MEDIUM

Absstract of: EP4807589A1

The present disclosure relates to the technical field of signal detection, and discloses a finger pressing state detection method, training method, electronic device, and medium, for improving generalization capability of finger pressing state detection. The method partially includes: pre-processing an ultrasonic echo signal to obtain model input data, where the ultrasonic echo signal is a corresponding ultrasonic echo signal after an ultrasonic array emits an ultrasonic signal to a pressing area; and inputting the model input data into a trained neural network model, so that the neural network model outputs a finger pressing state.

GAZE DETERMINATION USING ONE OR MORE NEURAL NETWORKS

Absstract of: US20260267406A1

0000 Apparatuses, systems, and techniques are presented to predict gaze of an observer. In at least one embodiment, a network is trained to predict a gaze of one or more users based, at least in part, on one or more gazes corresponding to objects not always visible to the one or more users.

SYSTEMS AND METHODS FOR TRAINING NEURAL NETWORKS WITH SPARSE DATA

Absstract of: US20260268137A1

A method, computer readable medium, and system are disclosed for training a neural network model. The method includes the step of selecting an input vector from a set of training data that includes input vectors and sparse target vectors, where each sparse target vector includes target data corresponding to a subset of samples within an output vector of the neural network model. The method also includes the steps of processing the input vector by the neural network model to produce output data for the samples within the output vector and adjusting parameter values of the neural network model to reduce differences between the output vector and the sparse target vector for the subset of the samples.

SPATIO-TEMPORAL GRAPH NEURAL NETWORK-BASED PHOTOVOLTAIC POWER GENERATION CAPABILITY PREDICTION METHOD AND SYSTEM, AND MEDIUM

Absstract of: WO2026184237A1

Disclosed are a spatio-temporal graph neural network-based photovoltaic power generation capability prediction method and system, and a medium. The method comprises: preprocessing historical data of a power system; dividing a plurality of power plants into a plurality of clusters, abstracting the power system into a topology graph, each cluster being regarded as a node of the topology graph; using a temporal self-attention network model to learn a temporal feature of the topological graph; using a spatial graph convolutional network model to learn a spatial feature of the topology graph; using a factorized structure to stack a temporal self-attention network and a spatial graph convolutional network; and using a fully connected layer to perform prediction. By modeling a photovoltaic power generation system as a graph model, the present invention can effectively capture spatial relationships and time dependencies between different power plants, thereby achieving more efficient data processing. This not only reduces redundant information, but also optimizes data flows, reducing storage resource requirements by reducing direct operations on raw data, and thereby improving hardware processing speeds.

METHOD FOR DIAGNOSING CARDIOVASCULAR DISEASE AND DEVICE USING THE SAME

Absstract of: US20260269067A1

Disclosed are a method for diagnosing cardiovascular disease and a device using the same. A control method of a diagnostic device according to one embodiment may comprise: obtaining a retinal image of a subject; and obtaining cardiovascular disease diagnostic information about the subject by using a machine learning model based on the retinal image, wherein the machine learning model includes a first model and a second model, wherein the first model is a neural network model, and wherein the second model may be a regression-based machine learning model.

ELECTRONIC DEVICE FOR PERFORMING VISION PERCEPTION FROM IMAGE ACQUIRED USING META LENS, AND OPERATING METHOD THEREOF

Absstract of: US20260270579A1

0000 An electronic device includes: a meta lens including pillars or pins provided at a surface of the meta lens and having at least one of different shapes, heights and widths; and an image sensor configured to receive phase-modulated light reflected from an object and transmitted by the meta lens, and obtain a coded image by converting the phase-modulated light into an electrical signal; and at least one processor configured to input the coded image into an artificial intelligence model, and obtain a label indicating a perception result of an object through inference using the artificial intelligence model, wherein the artificial intelligence model is a neural network model trained to obtain a simulation image by inputting an RGB image into a model reflecting optical characteristics of the meta lens, and output, as the perception result of the simulation image, a label indicating ground truth of the RGB image that was input.

METHOD AND APPARATUS FOR RECOGNISING CONSTRUCTION PRODUCTS AND/OR PROCESSES AT A CONSTRUCTION SITE

Absstract of: US20260268659A1

The present invention relates to a method and an apparatus for recognizing construction products and/or construction processes at a construction site, wherein, by means of a sensor system, a construction product and/or process at the construction site is detected and a product- and/or process-specific sensor data record is provided which is evaluated by means of an artificial neural network to identify and/or characterize the construction product and/or process. It is proposed that the recognition of the construction products and/or processes is no longer carried out by a central artificial neural network and instead, for this, a plurality of separate artificial neural networks are used which have been trained differently from one another and only for different subsets of the construction products and/or construction processes provided at the construction site.

METHODS AND SYSTEMS FOR IMPLEMENTING SECURE AND TRUSTWORTHY ARTIFICIAL INTELLIGENCE

Absstract of: US20260268024A1

Provided herein are systems, methods, computer-readable media, and techniques for providing trusted artificial intelligence (AI) using Fully Homomorphic Encryption (FHE), comprising: (A) providing a deep neural network (DNN)-based model with modified architecture, wherein the modified architecture at least (i) uses a Gaussian function as an activation function and (ii) removes one or more pooling layers; (B) obtaining encrypted data, wherein the encrypted data are generated by applying the FHE to plaintext data; and (C) generating an inference with the DNN-based model based on the encrypted data. Further provided herein are systems, methods, computer-readable media, and techniques for providing trusted AI using stochastic computing, using noise based computing, using an artificial immune system, and with secure multi-party computation (SMPC) based at least in part on watermarking.

ACCELERATED PHASE SPACE EXPLORATION OF MOLECULAR SYSTEMS USING NEURAL NETWORKS

Absstract of: US20260269024A1

A computer-implemented method for accelerating the simulation of a molecular system is provided. The method includes learning a first set of parameters by executing a Hamiltonian Monte Carlo method on the molecular system based on a set of initial conditions. A simulation of the molecular system is executed based on the first set of parameters. Thermodynamics expectation values of the molecular system are predicted based on the executed simulation. The predicted thermodynamics expectation values are provided for a downstream machine learning (ML) task. The method has applications including, but not limited to use cases in artificial intelligence (AI), drug development, medical diagnostics/applications, healthcare, material design catalyst design and high performance computing, to optimize predictions, improve model (e.g., neural network) performance or support decision making.

Computation Device, Updating Method, and Environment Recognition Device

Absstract of: US20260268143A1

A computation device comprising an inference unit that uses an inference model, which is a neural network model, and outputs an inference result corresponding to input data, and an updating unit that updates a weight of the inference model based on the inference result, and spatial information indicating at least a part of an area in the input data.

Systems and Methods for Fast Object Detection in Compressed Domains

Absstract of: US20260268513A1

0000 Methods and systems for performing detections at native resolution in high resolution video streams without requiring reconstruction of each frame comprises a baseline detector for performing object detections on key frames such as I frames in H.264 or H.265 video together with a shift lightweight neural network for calculating, from accumulated motion vectors of intermediate frames such as B frames and P frames, shift in the object detections and also together with a refine lightweight neural network responsive to the calculations from the shift network and accumulated frame residuals of the intermediate frames to determine object detections in the intermediate frames.

A METHOD OF TRAINING AN ARTIFICIAL DEEP NEURAL NETWORK FOR ESTIMATION OF HEMODYNAMIC PARAMETER, A METHOD OF ESTIMATION OF HEMODYNAMIC PARAMETER, COMPUTER PROGRAM PRODUCTS AND COMPUTER SYSTEMS

Absstract of: US20260268486A1

A method of training of an artificial deep neural network (ADNN) for estimation of a hemodynamic parameter from a geometry of a blood vessel tree comprising a step of obtaining of a set of geometries of vessel trees, according to the invention is realized with the ADNN having an architecture adapted for point cloud processing with distance based point grouping. The distance is defined as geodesic distance along the blood vessel tree. A method of estimation of hemodynamic parameters from a geometry of a blood vessel tree using an ADNN, according to the invention, involves using ADNN adapted for point cloud processing with geodesic distance based point grouping. The invention concerns also a computer program products comprising a set of instruction that, when run on a computing system, cause it to realize the methods according to the invention. The invention concerns also a computer system adapted to realize methods according to the invention.

NEURAL NETWORKS FOR INDICATING PRIORITY OF CONNECTION REQUESTS

Absstract of: US20260270721A1

0000 Apparatuses, systems, and techniques are described to predict priority of a connection request (e.g., a next connection request). For example, processing circuitry uses one or more neural networks to indicate priority of one or more next connection requests within one or more radio access network (RAN) networks. A base station of a UE device can use, perform, or otherwise execute the one or more neural networks to generate a prediction of what a next connection request will be.

DISTRIBUTED COMPUTATION OF DATA SEQUENCE PREDICTIONS USING NEURAL NETWORKS

Absstract of: WO2026188007A1

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for distributed computation of data sequence predictions using neural networks. One of the methods includes: receiving data defining an input data sequence; processing the data to generate a prediction output characterizing the input data sequence, comprising: extracting a plurality of data subsequences from the input data sequence; transmitting each of the plurality of data subsequences from a host system to a respective computing unit from a plurality of computing units; and processing, by each of the plurality of computing units and at least partially in parallel, the data subsequence transmitted to the computing unit to generate a prediction output for the data subsequence transmitted to the computing unit, wherein the prediction outputs generated by the plurality of computing units for the plurality of data subsequences collectively define the prediction output for the input data sequence.

DOMAIN-SPECIFIC COMMUNICATIONS USING DIALOGUE AND MANAGEMENT AGENTS

Absstract of: WO2026188040A1

Systems and methods for effectively generating responses to user inputs during a domain-specific communication session using generative neural networks. In some implementations, during a given communication session, the system makes use of a domain-specific conversational agent and a management agent, each of which include respective generative neural networks, to generate and communicate management plans to the user for a particular domain. In some implementations, during a given communication session, the system can generate responses to any given user input based on which of multiple phases the communication session is in.

METHOD AND SYSTEM FOR NON-INVASIVE ANIMAL IDENTIFICATION USING CONVOLUTIONAL NEURAL NETWORKS

Absstract of: WO2026185133A1

The present disclosure pertains to the field of machine learning and computer vision, specifically focusing on non-invasive animal identification systems utilizing convolutional neural networks and attribution methods. The present disclosure further pertains to data processing apparatus capable of carrying out the methods or housing information used in or generated from the methods.

SYSTEMS AND METHODS FOR MULTI-TIER ORCHESTRATION OF ARTIFICIAL INTELLIGENCE INFERENCE

Absstract of: US20260268182A1

0000 A distributed artificial intelligence inference system comprising a hierarchical multi-tier architecture with device, edge, metro, and data center tiers. A workload distribution controller computes a composite routing score for each inference task based on computational complexity, latency requirement, privacy classification, data volume, device resource state, and network conditions, and selects a target processing tier accordingly. A privacy classification engine assigns sensitivity levels to data elements, constraining eligible processing tiers. Multiple inference pipelines are co-located on shared edge nodes and exchange data via local shared memory. A model-architecture-specific state compression engine compresses session state using methods selected based on whether the inference pipeline employs a state-space model, transformer, or convolutional neural network architecture. A predictive routing engine computes destination confidence scores and proactively transfers compressed session state to predicted destination edge nodes to maintain session continuity during user mobility.

METHOD OF FINE-TUNING A NEURAL NETWORK USING VR AVATARS

Nº publicación: WO2026187248A1 10/09/2026

Applicant:

AKTSIONERNOYE OBSHCHESTVO SINTEZ [RU]
\u0410\u041A\u0426\u0418\u041E\u041D\u0415\u0420\u041D\u041E\u0415 \u041E\u0411\u0429\u0415\u0421\u0422\u0412\u041E \"\u0421\u0418\u041D\u0422\u0415\u0417\"

Absstract of: WO2026187248A1

The invention relates to the field of computing and artificial intelligence, and more particularly to methods of training neural networks. The present method includes digitizing the movements of an athlete in a video stream by analyzing a sequence of frames. A three-dimensional model of the athlete's movements is created which contains a parameterized skeleton comprising keypoint coordinates and joint angles, video files are generated in which the background and the appearance of the digitized athlete are altered, and a training database is created which contains rendered videos with modified appearances and backgrounds, and further contains labelling files containing information about movement parameters, environmental conditions, and characteristics of actions. A neural network is then fine-tuned by uploading the rendered videos and the labelling files, and tuning the architecture of a model analyzing the movement in the rendered videos taking into account the labelling data. The technical result consists in improving the quality of neural network training by using synthetically created data containing variations in movements, appearance and environment.