Alerta

GENERATING AUDIO USING AUTO-REGRESSIVE GENERATIVE NEURAL NETWORKS

Resumen de: EP4765100A2

0001 Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating a prediction of an audio signal. One of the methods includes receiving a request to generate an audio signal; obtaining a semantic representation of the audio signal; generating, using one or more generative neural networks and conditioned on at least the semantic representation, an acoustic representation of the audio signal; and processing at least the acoustic representation using a decoder neural network to generate the prediction of the audio signal.

DATA-FREE POST-TRAINING QUANTIZATION METHOD AND APPARATUS, DEVICE, AND STORAGE MEDIUM

Resumen de: EP4764966A1

This application provides a data-free post-training quantization method and apparatus, a device, and a storage medium, and relates to the field of neural network technologies. The method includes: obtaining data distribution input by a user, where the data distribution is distribution to which an input activation value of each network layer of a floating-point model conforms; inputting random data into the floating-point model to obtain the input activation value of each network layer; performing statistical analysis on the input activation value of each network layer based on the data distribution, to obtain a data range of the input activation value of each network layer; determining a quantization parameter of the input activation value of each network layer based on endpoint values of the data range; and during inference by using the floating-point model, performing, by using the quantization parameter of the input activation value of each network layer, quantization processing on the input activation value generated during inference of each network layer. According to the solution of this application, quantization processing can be performed on an input activation value in a data-free manner.

DATA PROCESSING METHOD AND APPARATUS

Resumen de: EP4765041A1

A data processing method is provided. The method is applied to image processing and includes: obtaining first data collected by an image sensor; and obtaining spectral information based on the first data by using a neural network model, where the neural network model includes an attention module, and the attention module is configured to determine an attention matrix based on input data, and perform an attention operation based on the attention matrix, where the attention matrix is obtained by performing a first fusion operation on correlation information between different channels of the input data and correlation information of the channels. In this application, a degree of correlation between the different channels and a degree of correlation of the channels may be fused, so that the attention matrix can model both correlation and particularity between the different channels, thereby improving accuracy of spectral signal reconstruction.

NEURAL NETWORK PROCESSOR AND METHOD OF NEURAL NETWORK PROCESSING

Resumen de: US20260170291A1

A neural network processor is provided comprising a plurality of mutually succeeding neural network processor layers is provided. A neural network processor layer therein comprising a plurality of neural network processor elements (1) having a respective state register (2) for storing a state value (X) indicative for their state, as well as an additional state register (4) for storing a value (Q) of a state value change indicator that is indicative for a direction of a previous state change exceeding a threshold value. Neural network processor elements in a neural network processor layer are configured to selectively transmit differential event messages indicative for a change of their state, dependent both on the change of their state value and on the value of their state value change indicator.

METHOD AND APPARATUS WITH NEURAL NETWORK DATA PROCESSING

Resumen de: US20260170820A1

0000 A processor-implemented neural network data processing method includes: determining a total number of either one of a first feature value and values less than or equal to the first feature value, in feature data output from a layer of a neural network; determining a quantization parameter based on the determined number; quantizing the feature data based on the determined quantization parameter; and inputting the quantized feature data to a another layer of the neural network connected to the layer.

DEEP LEARNING-ENHANCED CHEMILUMINESCENCE VERTICAL FLOW ASSAY FOR HIGH-SENSITIVITY ANALYTE DETECTION

Resumen de: WO2026128643A1

A method of measuring analyte(s) or biomarker(s) in a sample uses a cartridge-based vertical flow assay with a sensing membrane located within one or more cartridges and populated with a plurality of spots containing one or more capture agent(s). A mixture of the sample and detection reagents along with a chemiluminescence (CL) reagent solution are input into the one or more cartridges and the sensing membrane is imaged with a reader device configured to obtain one or more CL images and/or CL signals for the plurality of spots. The one or more images and/or CL signals for the plurality of spots are processed with an algorithm including machine learning or one or more trained neural networks configured to generate one or more outputs that include a classification of the sample and/or a quantification of the amount or concentration of the analyte(s) or biomarker(s) in the sample.

A COMPUTER-IMPLEMENTED METHOD FOR OPTIMIZING A NEURAL NETWORK MODEL FOR 3D OBJECT DETECTION

Resumen de: WO2026127827A1

A computer-implemented method for optimizing a neural network model for three-dimensional (3D) object detection. The method comprises receiving a pretrained 3D object detection model with multiple neural network layers and computes a layer-wise sparsity allocation across the detection model based on a predefined computational constraint. The layer-wise sparsity allocation is transformed into a layer-wise pruning ratio for each layer using second-order Hessian-based rate-distortion analysis, where the pruning ratio minimizes distortion in detection outputs. The computed pruning ratios are applied to remove redundant weights from each layer of the model, producing a pruned, pretrained 3D object detection model. This method reduces computational complexity while maintaining detection accuracy, making it suitable for real-time 3D perception applications.

GENERATIVE NEURAL NETWORKS WITH EFFECTIVE AUDIO TOKEN PROCESSING

Resumen de: WO2026128699A1

Methods, systems, and apparatuses, including computer programs encoded on computer storage media, for processing a received network input that includes audio data using a generative neural network to generate an output sequence that represents a transcription of speech included in the audio data. Then processing the output sequence of output tokens to generate a speech recognition output. One of the described techniques include training the generative neural network to generate outputs that interleave audio and text tokens. Another of the described techniques includes receiving and generating audio at the same time step.

ADVERSARIAL EXAMPLE PURIFICATION METHOD BASED ON CONDITIONAL DIFFUSION MODEL

Resumen de: WO2026123388A1

Disclosed in the present application is an adversarial example purification method based on a conditional diffusion model, comprising the following steps: acquiring a clean example dataset containing clean examples; using a white-box attack algorithm to attack a classification model to generate an adversarial example for each clean example; pairing the clean examples and the adversarial examples in one-to-one correspondence to form a training dataset; acquiring a pre-trained Stable Diffusion model; designing a fine-tuning process and a fine-tuning loss function; S6, presetting the number of iterations and a batch size, and using the training dataset to fine-tune network parameters of a cross-attention layer in the Stable Diffusion model to obtain a fine-tuned conditional diffusion model for adversarial example purification. In the present invention, an adversarial example is inputted as condition information into a UNet neural network to guide model learning, enabling the model to learn features of an adversarial perturbation, and improving computational efficiency, robustness and stability.

SENTIMENT CLASSIFICATION METHOD AND SYSTEM FOR SOCIAL NETWORK DYNAMICS, DEVICE AND STORAGE MEDIUM

Resumen de: WO2026123426A1

A sentiment classification method and system for social network dynamics, a device, and a storage medium. The method comprises: preprocessing a text of social dynamics to obtain a preprocessed data set; on the basis of the data set, constructing a semantic graph comprising word nodes and social dynamics nodes; extracting associated information between the social dynamics on the basis of topic attributes of the social dynamics in the semantic graph and inter-user relationships of users who publish the social dynamics, and establishing a connection relationship between the social dynamics nodes on the basis of the associated information between the social dynamics, so as to obtain a multi-layer social dynamics graph comprising a semantic relationship and a social relationship; and inputting the multi-layer social dynamics graph into an integrated model for processing, to obtain a sentiment classification result of the social dynamics, wherein the integrated model is composed of a hyperbolic learning-based graph convolutional neural network and a large-scale pre-trained language model.

ELECTROCARDIOGRAM-SIGNAL-BASED ASSISTED IDENTIFICATION METHOD, APPARATUS AND SYSTEM FOR ATTENTION DEFICIT HYPERACTIVITY DISORDER

Resumen de: WO2026124091A1

An electrocardiogram-signal-based assisted identification method, apparatus and system for attention deficit hyperactivity disorder. The method comprises: (1) collecting electrocardiogram data of a subject for whom the risk level of attention deficit hyperactivity disorder is required to be assessed, and processing the electrocardiogram data; (2) using a one-dimensional convolutional neural network to perform deep feature extraction on the processed electrocardiogram data of said subject, generating a classification heatmap from a feature map of the convolutional neural network by means of Score-CAM, and extracting a time-domain feature, a frequency-domain feature and a local statistical feature from the generated classification heatmap; and (3) inputting the time-domain feature, the frequency-domain feature and the local statistical feature into a machine learning classifier for classification, so as to obtain an attention deficit hyperactivity disorder risk assessment result of said subject. Deep learning is combined with various machine learning methods, so that the classification performance is improved, and enhanced feature interpretability is also provided.

COMPUTER-READABLE RECORDING MEDIUM HAVING STORED THEREIN DATA SELECTION PROGRAM, INFORMATION PROCESSING APPARATUS, AND COMPUTER-IMPLEMENTED DATA SELECTION METHOD

Resumen de: US20260170336A1

A non-transitory computer-readable recording medium having stored therein a data selection program that causes a computer to execute a process including determining, for each of a plurality of pieces of graph structure data, a weight in data selection based on the number of adjacent nodes shared by two nodes for a set of the two nodes included in the graph structure data, and selecting training data to be used for training a neural network that predicts a presence or absence of a link between nodes included in the graph structure data input as input data from among the plurality of pieces of graph structure data based on the determined weight and information indicating a presence or absence of a link between nodes for each of one or more sets of two nodes.

TWO-STREAM LSTM METHOD FOR PREDICTING POWER LOAD OF PORT SHORE

Resumen de: US20260171803A1

The present disclosure relates to the technical field of electric power engineering, in particular to a two-stream Long Short-Term Memory (LSTM) method for predicting power load of port shore power. Loads. The method entails collecting longitudinal data to identify factors that affect power load data, performing correlation analysis to classify dominant and auxiliary features power loads; separately modeling the dominant and auxiliary features and generating a fusion feature map; constructing a Bayesian Optimization-Long Short-Term Memory (BO-LSTM) neural network, and inputting a fusion feature map into a two-stream time series learning module, extracting a deep representation of the dominant and auxiliary features, then introducing a channel attention mechanism is to weight a fusion feature vector, and outputting a power load prediction value by a residual correction module. The present disclosure significantly improves the prediction accuracy and robustness, and supports the real-time scheduling of the port shore power system.

HYDROGEL MICROSPHERE SORTING METHOD, ARTIFICIAL ORGAN PREPARATION METHOD, AND SYSTEM AND MEDIUM

Resumen de: WO2026124687A1

A hydrogel microsphere sorting method, an artificial organ preparation method, and a system and a medium. The sorting method comprises the following steps: collecting a hydrogel microsphere image in a microfluidic chip; analyzing the hydrogel microsphere image by means of a neural network, so as to obtain an analysis result; and on the basis of the analysis result, sorting hydrogel microspheres encapsulating target cells, wherein the neural network comprises a feature extraction layer and a multi-scale joint output head, and the hydrogel microsphere image is subjected to convolution by means of the feature extraction layer and is then processed by the multi-scale joint output head, so as to obtain the analysis result. Hydrogel microspheres encapsulating target cells are accurately identified and sorted by using droplet microfluidic technology, and standardized artificial organs can be prepared on the basis of the sorted hydrogel microspheres, thereby significantly improving the functional maturity and reliability of the artificial organs.

NEURAL NETWORK SCHEDULING MECHANISM

Resumen de: US20260170596A1

An apparatus to facilitate workload scheduling is disclosed. The apparatus includes one or more clients, one or more processing units to processes workloads received from the one or more clients, including hardware resources and scheduling logic to schedule direct access of the hardware resources to the one or more clients to process the workloads.

MULTI-AGENT REINFORCEMENT LEARNING WITH MATCHMAKING POLICIES

Resumen de: US20260170334A1

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a policy neural network having a plurality of policy parameters and used to select actions to be performed by an agent to control the agent to perform a particular task while interacting with one or more other agents in an environment. In one aspect, the method includes: maintaining data specifying a pool of candidate action selection policies; maintaining data specifying respective matchmaking policy; and training the policy neural network using a reinforcement learning technique to update the policy parameters. The policy parameters define policies to be used in controlling the agent to perform the particular task.

SEGMENTATION USING AN UNSUPERVISED NEURAL NETWORK TRAINING TECHNIQUE

Resumen de: US20260170658A1

Systems and methods to detect one or more segments of one or more objects within one or more images based, at least in part, on a neural network trained in an unsupervised manner to infer the one or more segments. Systems and methods to help train one or more neural networks to detect one or more segments of one or more objects within one or more images in an unsupervised manner.

INFRINGEMENT DETECTION METHOD, DEVICE AND SYSTEM

Resumen de: EP4760634A2

A method for detecting an infringement by vehicle operator is described. The method comprises detecting a vehicle; receiving one or more image of at least a part of the vehicle operator; automatically analysing with a neural network the one or more captured received image to detect an infringing act; and providing the one or more captured received images comprising the detected infringing act to thereby detect the infringement. Also described are a system, a device, a computer system and a computer program product all for detecting an infringement by a vehicle operator. The device may comprise one or more flash for illuminating the vehicle or a part thereof with light at a narrow band and one or more camera comprising a narrow band filter that lets through only the wavelengths of light produced by the one or more flash.

TRAINING APPARATUS, ESTIMATION APPARATUS, TRAINING METHOD, ESTIMATION METHOD, AND PROGRAM

Resumen de: EP4760728A1

The present disclosure is designed to improve the accuracy of total energy estimation by neural network potential (NNP). An aspect of the present disclosure provides a learning apparatus including a control unit to carry out learning of an estimation model which is a mathematical model for energy estimation of an analysis target atom based on an evaluation target feature quantity indicating a first sum to a u-th sum, wherein the u-th sum is a sum of a three-dimensional (3D) wave function representing a u-th atomic orbital of the evaluation target atom located in a system including one or more atoms in descending order of occupied energy potential, and a 3D wave function of a u-th atomic orbital of each of other atoms in the system within a predetermined range of distances from the evaluation target in descending order of occupied energy potential, wherein in the learning, the estimation model is updated to reduce a difference between the sum of results of executing the estimation model for each atom in the system and the energy of the system estimated by density functional theory.

Configurable task prompts for neural network document transcription

Resumen de: GB2702481A

A processor comprises circuits causing a neural network 320 to generate a document transcription 354 of a document image 321 according to a configurable combination of annotation types 325 provided within input 302 to the neural network, wherein the document transcription comprises respective annotations of the annotation types for corresponding portions of content included in the document transcription. The document transcription may comprise a sequence of tokens, wherein the respective annotations correspond to description tokens in the sequence and the portions of content correspond to content tokens. The configurable combination of annotation types may comprise respective indicators that enable or disable individual annotation types of the plurality of annotation types. Annotation corresponding to disabled annotation types are not included in the document transcription. Annotation types may be specified in a multi-dimensional tuple and comprise: bounding boxes, semantic class labels, structured text, or plain text. The neural network may comprise a Vision Transformer (ViT) encoder 322, a compressor 324, and a decoder 326, wherein the document image 321 is input to the ViT encoder and the configurable combination of annotation types 325 is input to the decoder, and wherein the decoder outputs the document transcription. figure 3

CHIP SYSTEM, IMAGE PROCESSING METHOD AND ELECTRONIC DEVICE

Resumen de: EP4760630A1

Embodiments of this application provide a chip system, an image processing method, and an electronic device, and are applied to the field of chip image processing technologies. The chip system includes a graphics processing unit, a neural-network processing unit, and a scheduling circuit. The graphics processing unit is configured to output a first rendered image based on an image subtask. The scheduling circuit is configured to indicate, based on a scheduling subtask, the neural-network processing unit to obtain the first rendered image. The neural-network processing unit is configured to: based on a computing subtask, perform neural network algorithm processing on the first rendered image, and output a computing result. The scheduling circuit is further configured to indicate, based on the scheduling subtask, the graphics processing unit to obtain the computing result. The graphics processing unit is further configured to perform image processing on the computing result based on the image subtask to obtain a second rendered image. In embodiments of this application, image processing is performed based on multi-IP coordination to improve imaging quality. This manner reduces power consumption and processing complexity of the chip system, and is applicable to different chip architectures.

TRAINING NEURAL NETWORKS FOR SENSOR DATA REPRESENTATION LEARNING THROUGH FUTURE FRAME PREDICTION

Resumen de: EP4760597A1

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for obtaining respective sensor data captured by one or more sensors of an autonomous vehicle at each of a sequence of time steps, the sequence of time steps comprising one or more context time steps followed by one or more prediction time steps; generating respective ground truth birds-eye-view (BEV) representations of the respective sensor data for each of the prediction time steps; for each prediction time step, processing the respective sensor data at one or more preceding time steps in the sequence using a future prediction neural network to generate a predicted BEV representation for the prediction time step; and training the future prediction neural network based on, for each prediction time step, an error between the ground truth BEV representation for the prediction time step and the predicted BEV representation for the prediction time step.

Radio frequency digital signal classification

Resumen de: GB2702430A

The invention relates to the field of Radio Frequency (RF) signal classification and recognition, specifically a method of training a Convolutional Neural Network (CNN) and applying the trained CNN for the classification or identification of digitally modulated RF signals. The training comprises the steps of providing time-based training signals, generating IQ representations of those signals, applying k-means clustering to the IQ representations and generating a vector diagram by preserving timing indices of the IQ representations. A combined k-means cluster and vector image is generated (see Figure 3b). Once trained, the model is suitable for classifying digital signals, particularly digitally modulated communication signals. Figs. 2b & 3b

COMPOSITIONS AND METHODS FOR CELLULAR PHENOTYPE ASSESSMENT OF A SAMPLE USING CONFINED VOLUME ARRAYS

Resumen de: US20260159870A1

Certain embodiments of the invention are directed to evaluating and identifying cells by recording and interpreting a time-dependent signal produced by unique cell respiration and permeability attributes of isolated viable cells. Some methods comprise dividing the sample into two or more sub-samples or sample portions, mixing each sub-sample or sample portion with one or more reagents and/or one or more reactants forming distinct sub-sample or sample portion mixtures, compartmentalizing each of the sub-sample or sample portion mixtures into a plurality of small volume compartments, monitoring characteristics of the small volume compartments over time and collecting compartment data, and transmitting the collected data to at least one neural network.

METHOD OF CONTROLLING FOR UNDESIRED FACTORS IN MACHINE LEARNING MODELS

Nº publicación: US20260162184A1 11/06/2026

Solicitante:

STATE FARM MUTUAL AUTOMOBILE INSURANCE CO [US]
State Farm Mutual Automobile Insurance Company

Resumen de: US20260162184A1

A method of training and using a machine learning model that controls for consideration of undesired factors which might otherwise be considered by the trained model during its subsequent analyses of new data. For example, the model may be a neural network trained on a set of training images to evaluate an insurance applicant based upon an image or audio data of the insurance applicant as part of an underwriting process to determine an appropriate life or health insurance premium. The model is trained to probabilistically correlate an aspect of the applicant's appearance with a personal and/or health-related characteristic. Any undesired factors, such as age, sex, ethnicity, and/or race, are identified for exclusion. The trained model receives the image (e.g., a “selfie”) of the insurance applicant, analyzes the image without considering the identified undesired factors, and suggests the appropriate insurance premium based only on the remaining desired factors.