Neural networks

METHOD FOR ATTENTION-DRIVEN TRAINING OF COMPUTER-IMPLEMENTED NEURAL NETWORKS USING DOMAIN KNOWLEDGE

Absstract of: EP4818980A1

0001 Die vorliegende Erfindung betrifft ein computerimplementiertes aufmerksamkeitsgesteuertes Trainingsverfahren zum Erhalt eines trainierten neuronalen Netzes für die Schädigungserkennung und/oder für die Erkennung mindestens eines Bereichs einer Rissspitze in einer Abbildung eines zu untersuchenden Körpers, ein computerimplementiertes Verfahren für die Schädigungserkennung in Bilddaten eines zu untersuchenden Körpers und dessen Verwendung, ein Prüfsystem, ein Computerprogrammprodukt sowie ein computerlesbares Speichermedium. Hierbei umfasst das Trainingsverfahren die Einbindung von Domänenwisssen zur Steuerung der Aufmerksamkeit während des Trainings.

CLINICAL SUPPORT SYSTEM AND ASSOCIATED COMPUTER-IMPLEMENTED METHODS

Absstract of: EP4819077A2

A clinical support system comprises a processor and a display component, wherein: the processor is configured to: receive image data, the image data representing an image of a plurality of cells obtained from a human or animal subject, the image data comprising a plurality of subsets of image data, each subset comprising data representing a portion of the image data corresponding to a respective cell of the plurality of cells; apply a trained deep learning neural network model to each subset of the image data, the deep learning neural network model comprising: a plurality of convolutional neural network layers each comprising a plurality of nodes; and a bottleneck layer comprising no more than ten nodes, wherein the processor is configured to apply the trained deep learning neural network model to each subset of the image data by applying the plurality of CNN layers, and subsequently applying the bottleneck layer, each node of the bottleneck layer of the machine-learning model configured to output a respective activation value for that subset of the image data; for each subset of the image data, derive a dataset comprising no more than three values, the values derived from the activation values of the nodes in the bottleneck layer; and generate instructions, which when executed by the display component of a clinical support system, cause the display component of the computer to display a plot in no more than three dimensions of the respective dataset of each subset of the ima

PERSISTENT-PHOTOCONDUCTIVITY-BASED BIO-INSPIRED SENSING-MEMORY-COMPUTING INTEGRATED PHOTOELECTRIC DETECTION SYSTEM FOR MOVING TARGET

Absstract of: WO2026199857A1

The present application relates to the fields of bio-inspired intelligent vision and brain-inspired computing. Disclosed is a persistent-photoconductivity-based bio-inspired sensing-memory-computing integrated photoelectric detection system for a moving target. The system comprises a synapse-like bio-inspired photodetector array and a brain-inspired memristor array, which are interconnected, wherein the synapse-like bio-inspired photodetector array applies a fixed bias voltage to a device in each pixel when detecting visual information of the moving target, so as to convert spatial illumination information corresponding to each pixel into a persistently decaying photocurrent, and the synapse-like bio-inspired photodetector array continuously receives, during the persistent decay of the photocurrent, a plurality of frames of illumination information of the moving target, and reads the photocurrent of each pixel after the last frame of illumination information is inputted, so as to determine a feature map that stores all historical moment information; and the brain-inspired memristor array executes deep-neural-network operations on the feature map on the basis of Ohm's law and Kirchhoff's laws, and outputs a current corresponding to a task result label, so as to identify the moving target. The present application can meet the requirements of deep neural networks and complex dynamic vision tasks.

SCALING ELEMENT-WISE OPERATIONS OF A NEURAL NETWORK

Absstract of: US20260300712A1

Neural networks, and in particular deep neural networks, are traditionally resource intensive as they require execution of a significant number of computations in order to generate an output. In an effort to accelerate neural networks, quantization has been employed in which the input/output of certain operations are scaled. However, the degree to which a neural network can be quantized is limited because current scaling methods, which do not ensure consistent scaling for operands, cannot be applied to the commonly used element-wise operations (e.g. addition and subtraction). The present disclosure provides scaling of element-wise operations in a neural network by using scale factors that are shared between operands, thus allowing for greater quantization of neural networks.

GRAPH NEURAL NETWORKS

Absstract of: US20260301174A1

0000 A method comprising: generating subgraphs graph based on classification scores corresponding to nodes and obtained by inputting input graph into a graph neural network, GNN, wherein each node in each input graph comprises a plurality of values corresponding respectively to a plurality of features; determining statistical measures of each feature's values in the subgraphs to generate statistical measures for each subgraph, and obtaining for each subgraph classification probabilities corresponding respectively to a plurality of classes; performing k-means clustering on the subgraphs based on the statistical measures to generate a plurality of clusters of subgraphs; generating, for each cluster of subgraphs, a set of surrogate models to approximate the classification probabilities, respectively, of the subgraphs in the cluster, in terms of the statistical measures; and determining, based on the surrogate models of each set of surrogate models, feature importance scores.

METHOD AND DATA PROCESSING SYSTEM FOR LOSSY IMAGE OR VIDEO ENCODING, TRANSMISSION AND DECODING

Absstract of: WO2026207349A1

A method for lossy image or video encoding and transmission, the method including receiving a first image and classification information associated with the first image at a first computer system; with a first neural network, producing a latent representation of the first image using the classification information; transmitting the latent representation of the first image to a second computer system.

NOISE REGULATED STOCHASTIC SNN TO SOLVE QUADRATIC UNCONSTRAINED BINARY OPTIMIZATION PROBLEMS

Absstract of: WO2026202550A1

Noise regulation in stochastic spiking neural network (SSNN) allows for a relatively optimal exploration of the space of configurations, looking for the satisfiability of all the constraints; if applied discontinuously, it can also force the system to leap to a new random configuration effectively causing a restart. Embodiments use of noise regulation with an SSNN representing a QUBO problem to perform a parameter search to solve that QUBO problem.

SYSTEM AND METHOD FOR MULTI-LAYERED GRAPH INTEGRATION WITH DOMAIN-SPECIFIC AI AGENTS FOR DISPARATE DATA SOURCES

Absstract of: WO2026207116A1

A computer-implemented system integrates data from disparate sources into a multi¬ layered graph comprising domain-specific layers and an entity layer. The system includes an entity resolution engine configured to reconcile entity references by performing hash-based lookups for definitive identifiers and vector-based similarity searches for partial or ambiguous data. An ingestion pipeline parses incoming data, extracts attributes using neural network models, and updates the graph with new or matched nodes. Domain- specific Al agents, each associated with a respective layer, retrieve relevant data via retrieval-augmented generation (RAG) subsystems and generate insights using fine-tuned large language models (LLMs). A coordinating Al agent receives cross-layer queries, distributes sub-queries to domain-specific agents, and synthesizes a unified response by referencing the entity layer.

TUNING SWITCH PERFORMANCE PARAMETERS USING NEURAL NETWORKS

Absstract of: WO2026206395A1

Devices, computer-readable storage mediums, and methods tune performance parameters using one or more artificial intelligence models by monitoring data packets being received by a device, the data packets including one or more data characteristics, predicting, using the one or more artificial intelligence models, future data traffic patterns of the data packets, the predicting including categorizing the one or more data characteristics of the data packets, identifying, based on the predicted future data traffic patterns, one or more tuning parameters to improve bandwidth or reduce latency of the future data traffic patterns, and modifying one or more configurable parameters of the device to include the one or more tuning parameters.

METHOD AND SYSTEM FOR DETECTING AN OBJECT IN AN IMAGE

Absstract of: WO2026201760A1

The invention relates to a computer-implemented method for detecting an object in a digital image by means of a client-server system, wherein the following steps are carried out: a) providing a particular set of local training images to one of the at least two clients, wherein each local training image from the set comprises an image of the object, and the image includes a permissible or an impermissible representation of the object, which is characterised as such, b) generating and training a particular local embedding model on the basis of artificial intelligence for a feature of the object, by means of the particular client, from the particular set of local training images using an encoder, provided in each case to the particular client, in the form of a neural network having input nodes, intermediate layers and output nodes, wherein the number of output nodes is smaller than the number of input nodes, and the embedding model is formed by the output nodes of the encoder, c) transmitting the particular local embedding model from the particular client to the server, and aggregating the respective local embedding models into a global embedding model, and distributing the global embedding model to at least one of the at least two clients, d) capturing a current image of a produced product, which comprises a representation of the object, e) carrying out a feature analysis by applying the particular local embedding model, and detecting the object in the current image when a corr

DISTRIBUTION SYSTEM POWER-FLOW SOLUTION BY HIERARCHICAL ARTIFICIAL NEURAL NETWORKS STRUCTURE

Absstract of: US20260300691A1

Disclosed herein is a method for solving a power flow problem in at least one system. The method includes dividing, by at least one processor, the system into a plurality of clusters, where each of the plurality of clusters has a modular architecture, constructing, by the at least one processor, a hierarchical artificial neural network (ANN) for each of the plurality of clusters, thereby generating a plurality of ANNs, and determining, by the at least one processor by the plurality of ANNs, at least one solution to a power flow problem for (i) each of the plurality of ANNs, and (2) the hierarchical ANN as a whole, where each of the plurality of ANNs is organized hierarchically such that data from at least one lower-level layer is fed into at least one upper layer in accordance with an electric correlation between at least one cluster in the plurality of clusters.

NEURAL NETWORK MODEL INFERENCE METHOD AND COMPUTER DEVICE

Absstract of: WO2026199919A1

The present description provides a neural network model inference method and a computer device. A data owner receives N ciphertext neural network models sent by a model owner. The N ciphertext neural network models are obtained by performing obfuscation processing on a first neural network model. A true neural network model comprises a first linear layer, and N-1 fake neural network models obtained by means of obfuscation comprise second linear layers, wherein parameters of the second linear layers are 0. The first linear layer and the N-1 second linear layers are respectively encrypted on the basis of N key shares, wherein the N key shares are generated on the basis of a distributed encryption protocol supporting homomorphic operations. The data owner uses the N ciphertext neural network models to perform inference on target data to obtain the sum of N ciphertext outputs, thereby obtaining an inference result for the target data on the basis of the sum of the N ciphertext outputs. By using the N key shares for encryption and generating the fake neural network models for model obfuscation, the security of the models is improved.

NEURAL NETWORK MODEL INFERENCE METHOD AND COMPUTER DEVICE

Absstract of: WO2026199920A1

A neural network model inference method and a computer device. A data party first receives an encrypted neural network model sent by a model party. A first linear layer of the encrypted neural network model corresponds to a first ciphertext linear layer and N-1 second ciphertext linear layers, the first ciphertext linear layer being obtained by directly encrypting the first linear layer, and the second ciphertext linear layer being obtained by encrypting a linear layer with all zero parameters. The first ciphertext linear layer and the second ciphertext linear layer are respectively encrypted using N key shares, the N key shares being generated on the basis of a distributed encryption protocol supporting homomorphic operations. The data party uses the encrypted neural network model to perform target data inference to obtain an intermediate result output by another layer preceding the first linear layer, and inputs the intermediate result into the first linear layer and the N-1 second linear layers, respectively, to obtain N outputs respectively corresponding to N ciphertext linear layers. An inference result of the target data is obtained according to a sum of the N outputs.

DETECTION AND CLASSIFICATION OF TRAFFIC SIGNS USING CAMERA-RADAR FUSION

Absstract of: US20260299088A1

0000 The disclosed systems and techniques facilitate efficient detection and classification of traffic signs in driving environments. The disclosed techniques include, obtaining, using a sensing system of a vehicle a first set of perspective camera images of an environment and a second set of radar images of the environment. The techniques further include generating, using a first neural network, one or more camera features characterizing the first set of images, generating, using a second neural network, one or more radar features characterizing the second set of images, and processing the one or more camera features and the one or more radar features to obtain an identification of one or more traffic signs in the environment.

DETECTION OF UNSAFE CABIN CONDITIONS IN AUTONOMOUS VEHICLES

Absstract of: US20260296364A1

0000 Devices, systems and processes for the detection of unsafe cabin conditions that provides a safer passenger experience in autonomous vehicles are described. One example method for enhancing passenger safety includes capturing at least a set of images of one or more passengers in the vehicle, determining, based on the set of images, the occurrence of an unsafe activity in an interior of the vehicle, performing, using a neural network, a classification of the unsafe activity, and performing, based on the classification, one or more responsive actions.

METHOD FOR RECOGNIZING SPACES TO BE OPTIMIZED IN BUILT ENVIRONMENT ON BASIS OF MORPHOLOGICAL HIERARCHY MODEL

Absstract of: WO2026199950A1

A method for recognizing spaces to be optimized in a built environment on the basis of a morphological hierarchy model, comprising collecting and preprocessing data of a built environment, preliminarily performing hierarchical classification of the morphologies of the built environment, correcting the hierarchical classification of the morphologies of the built environment, preliminarily recognizing spaces to be optimized, performing hierarchical recognition of said spaces, and performing recognition result interaction and feedback. According to the method, data collection is performed by means of a laser radar system, an aerial photography unmanned aerial vehicle, a street view collection vehicle, and a low earth orbit satellite, and recognition and display of spaces to be optimized in a built environment are performed on the basis of multi-task Bayesian federated learning, an XGBoost model, a vision transformer, and a convolutional neural network. The described method can facilitate rapid recognition of spaces to be optimized in a built environment in the field of urban planning, and improve the rationality and accuracy of recognition results of said spaces in the built environment by means of morphological hierarchical classification.

Ocular system for deception detection

Absstract of: AU2026233962A1

A method of deception detection, assessing operational risk or optimizing learning is based upon ocular information of a subject by providing a video camera configured to record a close-up view of a subject's eye. The ocular information is processed to identify changes in ocular signals of the subject through the use of convolutional neural networks. Changes in ocular signals are evaluated from the convolutional neural networks by a machine learning algorithm. Results can then be presented in regards to the level of deception, the level of operational risk or the best way to optimize learning. The methods are facilitated by identifying at least one predictive distortion identified in the stroma capturable solely with a visible-spectrum camera correlating to a predicted response in the iris musculature. 20 ep e p

DISTANCE DETERMINATIONS USING ONE OR MORE NEURAL NETWORKS

Absstract of: US20260300718A1

0000 Apparatuses, systems, and techniques are presented to determine distance for one or more objects. In at least one embodiment, a disparity network is trained to determine distance data from input stereoscopic images using a loss function that includes at least one of a gradient loss term and an occlusion loss term.

NEURAL NETWORK TRAINING TECHNIQUE

Absstract of: US20260300436A1

0000 Apparatuses, systems, and techniques to train neural networks to perform image processing tasks. In at least one embodiment, one or more second neural networks are used to train one or more first neural networks based, at least in part, on a first object type in one or more images and a second object type in the one or more images, in parallel.

SYSTEM AND METHOD FOR HUMAN ACTIVITY RECOGNITION

Absstract of: US20260300738A1

0000 In aspects, the present approaches allow neural networks to be taught to understand patterns of human behavior without the need of expert data labeling or laboratory studies. First and second neural networks are trained to understand these patterns without labeling. Once trained, the neural networks can be deployed with a trained classifier to determine or classify human activity based upon received sensor inputs.

SELECTION-INFERENCE NEURAL NETWORK SYSTEMS

Absstract of: US20260296464A1

0000 Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating a response to a query input using a selection-inference neural network.

AUTOMATIC BALL MACHINE APPARATUS UTILIZING PLAYER IDENTIFICATION AND PLAYER TRACKING

Absstract of: US20260295361A1

0000 A ball machine comprising an imaging system to capture image data and a processor configured to, for a frame of the image data, analyze the image data using a neural network to detect a plurality of persons, determine a coordinate position on a playing surface of each of the plurality of detected persons, extract features of each of the plurality of detected persons, generate a first set of feature vectors corresponding to the plurality of detected persons, associate a first feature vector to the coordinate position on the playing surface of a first detected person to generate a first unique identifier, associate a second feature vector to the coordinate position on the playing surface of a second detected person to generate a second unique identifier, and control the ball machine to launch balls based on first settings corresponding to the first unique identifier and second settings corresponding to the second unique identifier.

TEMPORAL IMAGE BLENDING USING ONE OR MORE NEURAL NETWORKS

Absstract of: US20260301117A1

Apparatuses, systems, and techniques are presented to reconstruct one or more images. In at least one embodiment, one or more objects in an image are caused to be generated based, at least in part, on applying one or more offsets to a motion of the one or more objects relative to one or more prior images.

INFORMATION PROCESSING APPARATUS, INFORMATION PROCESSING METHOD, AND PROGRAM

Absstract of: EP4814884A1

0001 An information processing apparatus classifies cells in an image of a tumor microenvironment obtained from a patient into a plurality of cell groups, extracts positions, in the image, of the cells included in the plurality of cell groups, creates a graph by assigning nodes to the cells included in the plurality of cell groups based on the extracted positions, connecting the nodes with edges, assigning to each of the nodes a feature quantity indicating a feature of the corresponding cell, and assigning to each of the edges a feature quantity indicating a relationship between two cells corresponding to two nodes connected by the edge, and models, using a graph neural network (GNN), spatial relationships between cells in the graph based on the assigned feature quantities of the nodes and the edges.

SYSTEM AND METHOD FOR PLAYER REIDENTIFICATION IN BROADCAST VIDEO

Nº publicación: EP4815476A2 30/09/2026

Applicant:

STATS LLC [US]
STATS LLC

Absstract of: EP4815476A2

0001 A system and method of re-identifying players in a broadcast video feed are provided herein. A computing system retrieves a broadcast video feed for a sporting event. The broadcast video feed includes a plurality of video frames. The computing system generates a plurality of tracks based on the plurality of video frames. Each track includes a plurality of image patches associated with at least one player. Each image patch of the plurality of image patches is a subset of the corresponding frame of the plurality of video frames. For each track, the computing system generates a gallery of image patches. A jersey number of each player is visible in each image patch of the gallery. The computing system matches, via a convolutional autoencoder, tracks across galleries. The computing system measures, via a neural network, a similarity score for each matched track and associates two tracks based on the measured similarity.