Alerta

PROCESSING IMAGES USING MIXTURE OF EXPERTS

Resumen de: US20260340783A1

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating predictions about images. One of the systems includes a neural network comprising a sequence of one or more network blocks that are each configured to perform operations comprising: obtaining a block input that represents an intermediate representation of an input image; determining a plurality of patches of the block input or of an updated representation of the block input, wherein each patch comprises a different subset of elements of the block input or of the updated representation of the block input; assigning each patch to one or more respective expert modules of a plurality of expert modules of the network block; for each patch of the plurality of patches, processing the patch using the corresponding expert modules to generate respective module outputs; and generating a block output by combining the module outputs.

OPTIMIZING ALGORITHMS FOR TARGET PROCESSORS USING REPRESENTATION NEURAL NETWORKS

Resumen de: US20260339634A1

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for optimizing a target algorithm using a state representation neural network.

METHODS FOR COFFEE BEAN DEFECT DETECTION BASED ON SIAMESE NEURAL NETWORKS

Resumen de: US20260340809A1

Disclosed is a method for coffee bean defect detection based on a Siamese neural network. The method includes: obtaining image data of normal coffee beans and four types of defective coffee beans, performing a preprocessing operation on the image data, and dividing the preprocessed image data into a training set, a validation set, and a test set according to a preset ratio; constructing a coffee bean defect detection model based on a dual-branch Siamese neural network, and training the coffee bean defect detection model using the training set to obtain a similarity score; validating and adjusting the trained coffee bean defect detection model using the validation set by employing a mixed-precision training strategy and a dynamic learning rate adjustment technique; and performing defect detection inference on the validated and adjusted coffee bean defect detection model using the test set to complete the coffee bean defect detection.

TECHNIQUES TO PROCESS LAYERS OF A THREE-DIMENSIONAL IMAGE USING ONE OR MORE NEURAL NETWORKS

Resumen de: US20260340885A1

Apparatuses, systems, and techniques to identify one or more layers of a three-dimensional graphical image to generate a two-dimensional representation. In at least one embodiment, one or more layers of a three-dimensional graphical image are identified to generate one or more two-dimensional representations.

Reinforcement learning of tactile grasp policies

Resumen de: AU2026253872A1

One or more processors comprising: circuitry to enable one or more robotic devices to grasp one or more objects by: identifying, in one or more images depicting a scene that includes the one or more objects, one or more bounding volumes enclosing the one or more objects; using one or more neural networks to identify a target pose of one or more robotic devices to interact with the one or more objects based, at least in part, on one or more measurements from one or more tactile sensors attached to the one or more robotic devices and the bounding volumes enclosing the one or more objects identified in the one or more images; and causing the one or more robotic devices to grasp the one or more objects based, at least in part, on the identified target pose. measurements from one or more tactile sensors attached to the one or more robotic devices ct c t m e a s u r e m e n t s f r o m o n e o r m o r e t a c t i l e s e n s o r s a t t a c h e d t o t h e o n e o r m o r e r o b o t i c d e v i c e s

Interactive Image Generation Method

Resumen de: US20260341003A1

An interactive image generation method receives a communication that requests that a described imaginative scenario is generated that is to be embodied by at least one image. The description of the imaginative scenario is interpreted by applying attentions in association with applying a trained neural network that is trained using a training set comprising syntactical elements and images. Pixels are generated in accordance with learned training correspondences and the imaginative scenario by applying the trained neural network. The pixels may be further generated in accordance with an inference of a user preference that is based on usage behaviors that occur prior to the communication. An image that includes the generated pixels is provided to a user.

ONLINE EVALUATION METHOD FOR ERROR OF CAPACITIVE VOLTAGE TRANSFORMER

Resumen de: WO2026223030A1

The present invention relates to the technical field of online monitoring of capacitive voltage transformers (CVT), and in particular to an online evaluation method for an error of a CVT, comprising the following steps: step 1: collecting a large amount of historical operating data having error labels of a CVT; step 2: selecting suitable main features affecting errors; step 3: preprocessing the historical data and operating errors and constructing a data set; step 4: establishing an operating error evaluation model combining a variational autoencoder and a convolutional neural network; step 5: using a training set and a test set to train the operating error evaluation model; and step 6: when the trained model is put into use, subjecting online monitoring data of a CVT of the same model to the same preprocessing, and then inputting the preprocessed data into the trained operating error evaluation model to obtain an operating error. Online analysis and evaluation of the operating error of the CVT is realized; and further, building the evaluation model combining the variational autoencoder and the convolutional neural network improves the analysis accuracy.

INDICATING MODEL PARAMETERS FOR CONTINUAL LEARNING OF COMMUNICATIONS NETWORK MODELS

Resumen de: WO2026224600A1

Various aspects of the present disclosure relate to adapting (or updating) an artificial intelligence/machine learning (AI/ML) model (e.g., a deep neural network) to unknown/new target domains while minimizing any loss of knowledge of previous domains within which the model was deployed or adapted. For example, an adaptation procedure may determine or identify knowledge intensive (e.g., important) parameters of the model and adapt the model to a new domain while minimizing changes to values of the knowledge intensive parameters. The procedure may determine a metric for each parameter of the model (e.g., a knowledge coefficient or importance metric) and minimize changes to values of any parameters having relatively high importance metrics during adaptation/updating of the model.

Joint Training of Neural Networks using Multi-Scale Hard Example Mining

Resumen de: US20260340979A1

An example apparatus for mining multi-scale hard examples includes a convolutional neural network to receive a mini-batch of sample candidates and generate basic feature maps. The apparatus also includes a feature extractor and combiner to generate concatenated feature maps based on the basic feature maps and extract the concatenated feature maps for each of a plurality of received candidate boxes. The apparatus further includes a sample scorer and miner to score the candidate samples with multi-task loss scores and select candidate samples with multi-task loss scores exceeding a threshold score.

PREDICTION AND EARLY WARNING SYSTEM FOR WATER QUALITY AT MONITORING-SCARCE POINT BASED ON LIQUID GRAPH NEURAL NETWORK

Resumen de: US20260338753A1

A system for the water quality based on a liquid graph neural network includes watershed front-end monitoring devices, an edge computing apparatus and a central server; the apparatus constructs a directed initial watershed graph, historical flow velocities, a riverbed permeability coefficient and hydrological connectivity data, and dynamically corrects edge weights based on real-time flow velocity gradients and rainfall changes; and generates a dynamic adjacency matrix and outputs predicted values for water quality parameters and alarm levels within a future preset time domain at a data-scarce point by using the neural network; when the predicted value exceeds a preset threshold, the apparatus sends a hardware action instruction to an associated upstream node or downstream node to control an automatic sampler to perform encrypted sampling, and calls a local model and locally cached data to maintain continuous early warning when a response delay of the central server reaches a preset condition.

IMAGE PROCESSING USING NORMALIZATION LAYERS

Resumen de: US20260340387A1

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing images or features of images using an image classification system that includes a batch normalization layer. One of the systems includes a convolutional neural network configured to receive an input comprising an image or image features of the image and to generate a network output that includes respective scores for each object category in a set of object categories, the score for each object category representing a likelihood that that the image contains an image of an object belonging to the category, and the convolutional neural network comprising: a plurality of neural network layers, the plurality of neural network layers comprising a first convolutional neural network layer and a second neural network layer; and a batch normalization layer between the first convolutional neural network layer and the second neural network layer.

INTELLIGENT EMPIRICAL CORRECTION METHOD AND SYSTEM FOR NUMERICAL OCEAN CURRENT FORECASTING BASED ON MULTIMODAL NEURAL NETWORK

Resumen de: WO2026223124A1

The present invention relates to the interdisciplinary technical field of artificial intelligence and ocean forecasting. Disclosed are an intelligent empirical correction method and system for numerical ocean current forecasting based on a multimodal neural network. The method comprises the steps of: collecting observation data of an ocean current region to be forecasted and constructing a label data set; preprocessing the label data set to construct a feature data set; on the basis of the label data set and the feature data set, using a multimodal neural network to construct a correction model; and on the basis of the correction model, completing the correction of ocean current forecasting. The present invention has the capability of adaptively capturing patterns among multi-source heterogeneous data, and improves an ocean current correction effect, thereby improving the accuracy of ocean current forecasting, and providing more reliable forecasting data for marine economic development, marine research, and climate prediction.

Marine Detection System and Method

Resumen de: US20260341158A1

Marine detection system and method for detecting marine bodies (8). The system includes a first sensor (6) for detecting marine bodies in a first detection region. A second sensor (3) is provided for detecting marine bodies in a second detection region. The second sensor is a different type of sensor to the first sensor and the second detection region overlaps with or is adjacent to the first. A controller (5) is provided including a neural network for processing data from the first and second sensors, where the neural network is trained to determine the presence of a marine body (8) based on the sensor data characteristics, and a signal indicating the presence of the marine body is output based on the determination by the neural network.

TRAINING NEURAL NETWORKS WITH GRADIENT CHECKPOINTING

Resumen de: US20260340378A1

A data processing system for training a neural network. The data processing system comprises a storage circuitry configured to store checkpoint data for training the neural network and training configuration data indicative of a data partition for training the neural network. The data processing system also comprises a processing circuitry configured to perform a checkpoint operation for a layer of the neural network. The checkpoint operation comprises: obtaining input feature map data for a layer of the neural network, obtaining the training configuration data, selecting a portion of the input feature map data based on the training configuration data, processing the selected input feature map data in the layer of the neural network to obtain output feature map data, selecting a portion of the output feature map data based on the training configuration data, and storing, in the storage circuitry, checkpoint data derived from the selected output feature map data.

AI INFERENCE COMPILER AND RUNTIME TOOL CHAIN

Resumen de: US20260339628A1

Embodiments include systems and methods for processing sensor data and generating operational instructions of hardware of egos (e.g., autonomous vehicles, robots). The ego includes any number of machine-learning architectures, often neural network architectures, for processing sensor data and recognizing the environment around the ego and making decisions on the ego's behavior. The neural network architectures of the ego ingest sensor data and execute any number of operations related to a particular domain or task, such as object recognition or path planning, using the sensor data. A graph partitioner is trained to assign functions in the software of the neural networks and the sensor data to certain hardware processing units. Several compilers are used to generate the instructions based upon the assigned type of processing unit.

INDEPENDENT SPLIT MODEL INFERENCE IN SPLIT NEURAL NETWORK FOR ESTIMATING NETWORK PARAMETERS

Resumen de: US20260340342A1

A split neural network includes a tail network model that receives a first plurality of activations and a second plurality of activations at a cut layer of the split neural network, and that generates a model output in response to the first plurality of activations and the second plurality of activations; a head network model that receives a plurality of input feature values and generates the first plurality of activations in response to the plurality of input feature values and provides the first plurality of activations to the tail network model at the cut layer; and a translator model that receives the first plurality of activations, that generates estimated values of the second plurality of activations in response to the first plurality of activations, and that provides the estimated values of the second plurality of activations to the tail network model at the cut layer.

SELECTION-INFERENCE NEURAL NETWORK SYSTEMS

Resumen de: US20260340353A1

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.

METHOD AND DEVICE FOR GENERATING AN OUTSIDER PERSPECTIVE IMAGE AND METHOD OF TRAINING A NEURAL NETWORK

Resumen de: US20260340929A1

According to various embodiments, a computer-implemented method for generating an outsider perspective image may be provided. The method may include receiving a plurality of input images that capture surroundings of an object. The method may further include projecting each input image of the plurality of input images, onto a respective virtual surface of a set of virtual surfaces around the object to generate a set of surface images. The method may further include projecting each surface image of the set of surface images onto a common coordinate frame, to generate a transformed dataset. The method may further include generating, by a machine learning model, an outsider perspective image based on the transformed dataset.

PADDING FEATURE MAPS IN A NEURAL NETWORK

Resumen de: US20260340766A1

Technology is disclosed herein to execute an inference model by a processor which includes a reshape layer. In an implementation, the reshape layer of the inference model receives an output produced by a previous layer of the inference model and inserts padding into the output, then supplies the padded output as an input to a next layer of the inference model. In an implementation, the inference model includes a stitching layer at the beginning of the inference model and an un-stitch layer at the end of the model. The stitching layer of the inference model stitches together multiple input images into an image batch and supplies the image batch as an input to a subsequent layer. The un-stitch layer receives output from a penultimate layer of the inference model and unstitches the output to produce multiple output images corresponding to the multiple input images.

ENHANCED LIFE ASSESSMENT METHOD FOR HIGH-VOLTAGE CIRCUIT BREAKER

Resumen de: WO2026222905A1

The present invention is applicable to the technical field of life assessment, and provides an enhanced life assessment method for a high-voltage circuit breaker. The method comprises: constructing a dynamic evolution model for the wear rate of a contact material; constructing a dynamic prediction model for nozzle degradation; constructing a probability distribution of the remaining life of an insulating material; on the basis of the dynamic evolution model for the wear rate of the contact material, the dynamic prediction model for nozzle degradation, and the probability distribution of the remaining life of the insulating material, constructing a comprehensive life assessment index; inputting the comprehensive life assessment index into a preset hierarchical prediction architecture for assessment, wherein a first layer generates a basic life distribution by means of a Bayesian network, a second layer outputs a correction parameter of a posterior life distribution by means of a convolutional neural network, and a third layer optimizes the posterior distribution by means of a KL divergence minimization algorithm to obtain a probability density function and a confidence interval of the remaining electrical life; and dynamically correcting model parameters of the hierarchical prediction architecture. The present invention can reduce errors in life assessment under complex operating conditions.

METHOD AND SYSTEM FOR DETECTING SMART CONTRACT STATE MANIPULATION ATTACK BASED ON GRAPH NEURAL NETWORK, DEVICE, AND MEDIUM

Resumen de: EP4831848A1

Provided are a method and a system for detecting a smart contract state manipulation attack based on a graph neural network, a device, and a medium, belonging to the technical field of blockchain security analysis. The method includes: defining the type of the smart contract state manipulation attack, collecting attack smart contracts and normal smart contracts on a blockchain, and constructing a training data set; positioning a creator of a target smart contract, and constructing a historical external transaction graph; decompiling the target smart contract, and constructing an external call behavior graph; combining the historical external transaction graph with the external call behavior graph, and constructing a creation behavior graph of a smart contract; constructing a classifier of a state manipulation attack; and monitoring a smart contract created in real time, constructing the creation behavior graph, and inputting the creation behavior graph into the classifier to detect the state manipulation attack in real time. Blockchain nodes can be connected, a newly deployed smart contract is acquired in real time, the creation behavior graph of the smart contract is constructed, and the creation behavior graph is input into a graph classification model to detect a state manipulation attack smart contract.

PROCESSING IMAGES USING SELF-ATTENTION BASED NEURAL NETWORKS

Resumen de: EP4831877A2

0001 Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing images using self-attention based neural networks. One of the methods includes obtaining one or more images comprising a plurality of pixels; determining, for each image of the one or more images, a plurality of image patches of the image, wherein each image patch comprises a different subset of the pixels of the image; processing, for each image of the one or more images, the corresponding plurality of image patches to generate an input sequence comprising a respective input element at each of a plurality of input positions, wherein a plurality of the input elements correspond to respective different image patches; and processing the input sequences using a neural network to generate a network output that characterizes the one or more images, wherein the neural network comprises one or more self-attention neural network layers.

SYSTEM AND METHOD FOR MONITORING MIXED FLOWS OF PEOPLE AND VEHICLES IN RURAL AND NATURAL ENVIRONMENTS

Resumen de: EP4831881A1

System and method for monitoring mixed flows of people and vehicles in rural and natural environments. The system comprises person and vehicle detection devices (2) equipped with a millimetre-wave radar (5) and a data processing unit (8) the purpose of which is to:- classify the moving point cloud (10) of each radar scan using a previously trained artificial neural network (18) that receives as input the distance (12), the azimuth angle (13), the elevation angle (14), the Doppler velocity (15), and the signal-to-noise ratio (16) of each point (11) in the cloud of moving points (10), to yield a list (20) of targets classified by type (23) as a person (24) or a specific vehicle (28, 29, 30) from a list of vehicles (25),- determine a trajectory (27) for each target (21),- count each type (23) of target the trajectory (27) of which crosses a counting line (33).A central server (3) monitors the flows of each type (23) of target using the count information (42).

DATA PROCESSING METHOD AND RELATED DEVICE

Resumen de: EP4831876A1

Embodiments of this application provide a data processing method and a related device. The method includes: step 1: obtaining first request information; step 2: obtaining, based on a first agent, first response information corresponding to the first request information; step 3: obtaining evaluation information of the first response information based on a second agent; step 4: optimizing the first response information based on the first agent and the evaluation information to obtain second response information; and step 5: training a neural network based on the first request information and the second response information. To be specific, the first agent may perform optimization based on the evaluation information provided by the second agent for the first response information, and train the neural network by using information in an optimization process, thereby improving an inference effect of the neural network. Alternatively, it is understood that evolution of the neural network can be sustained through interaction between the first agent and the second agent without relying fully on human supervision.

PATH LOSS DETERMINATION METHOD AND APPARATUS, STORAGE MEDIUM, AND ELECTRONIC DEVICE

Nº publicación: EP4831935A1 28/10/2026

Solicitante:

ZTE CORP [CN]
ZTE CORPORATION

Resumen de: EP4831935A1

0001 A path loss determination method and apparatus, a storage medium, and an electronic device are provided. The method includes: determining (S202) a superimposed field image of a target grid scene on the basis of a first image, a second image, and an incident field image of the target grid scene; determining (S204) a first loss function on the basis of the incident field image and the superimposed field image, and determining (S204) a second loss function on the basis of the superimposed field image and drive test data of the target grid scene; determining (S206) a target loss function on the basis of the first loss function and the second loss function; and training (S208) a neural network model on the basis of the target loss function, so as to determine the path loss of the target grid scene by means of the trained neural network model.