Alerta

DYNAMICALLY DIVIDING ACTIVATIONS AND KERNELS FOR IMPROVING MEMORY EFFICIENCY

Resumen de: US20260260115A1

Embodiments are generally directed to dynamically dividing activations and kernels for improving memory efficiency. An embodiment of a method in a compute engine performing machine learning comprises: receiving, by a convolutional layer of a convolutional neural network (CNN) implemented on the compute engine, a plurality of activation groups contained in an input data, wherein the convolutional layer includes one or more kernel groups and the one or more kernel groups each include a plurality of kernels; determining a plurality of memory efficiency metrics based on the number of activation groups of the plurality of activation groups and the number of kernels of the plurality of kernels; selecting a first optimal number of activation groups and a second optimal number of kernels that are associated with an optimal memory efficiency metric in the plurality of memory efficiency metrics; and performing a convolutional operation on the input data based on the first optimal number and the second optimal number.

Computer Vision Based Two-Stage Surgical Phase Recognition Module

Resumen de: US20260260488A1

0000 A system and method of identifying an adverse event using a video of a surgery and a two-stage surgical phase recognition module. The method includes receiving, by the module, a video of the surgery, where the video comprises a sequence of video frames. The module comprises a first stage that includes a neural network and a second stage that includes a multi-stage temporal convolution network. The method includes extracting, using the first stage, visual information content of a single frame based on the single frame; identifying, using the second stage, surgical phases captured in the frames of the video based on the visual information content from the first stage; and identifying, using the identified surgical phases, an adverse event during the surgery. An adverse event includes the omission of a surgical phase and an injury to the patient. The identification can occur in real-time or near-real-time.

DEEP LEARNING SYSTEM

Resumen de: US20260260113A1

0000 A machine learning system is provided to enhance various aspects of machine learning models. In some aspects, a substantially photorealistic three-dimensional (3D) graphical model of an object is accessed and a set of training images of the 3D graphical mode are generated, the set of training images generated to add imperfections and degrade photorealistic quality of the training images. The set of training images are provided as training data to train an artificial neural network.

SYSTEM, METHOD, AND COMPUTER PROGRAM PRODUCT FOR MACHINE UNLEARNING ON IDENTITY GRAPH NEURAL NETWORKS

Resumen de: WO2025090089A1

Systems, methods, and computer program products for machine unlearning on identity graph neural networks may obtain an identity graph including a plurality of graphs, each graph including a plurality of edges and a plurality of nodes for the plurality of edges, and, in each graph, each edge and each node is associated with a same identity; apply at least one edge augmentation algorithm to the identity graph to make the identity graph a biconnected identity graph; split the biconnected identity graph into a plurality of biconnected components, such that there are no articulation points in each biconnected component; for each biconnected component, train a graph neural network that corresponds to that biconnected component to generate a graph embedding and a local minima; train an ensemble neural network on the graph embedding and the local minima of each biconnected component; and provide the trained ensemble neural network.

COMPUTER-IMPLEMENTED METHOD FOR GENERATING AN ORTHODONTIC TREATMENT PLAN

Resumen de: US20260253325A1

0000 The present application provides a computer-implemented method for generating an orthodontic treatment plan, the method comprises: obtaining a first and a second 3D digital models, where the first 3D digital model represents an initial tooth arrangement of a jaw/jaws, and the second 3D digital model represents a target tooth arrangement of the jaw/jaws; extracting features from the first 3D digital model using a trained feature extraction deep neural network; and generating an orthodontic treatment plan of the jaw/jaws using a trained multi-agent reinforcement learning based deep neural network based on the extracted features and the second 3D digital model, where in the multi-agent reinforcement learning based deep neural network, each tooth is taken as an agent, where the orthodontic treatment plan utilizes shell-shaped tooth repositioners and comprises a series of successive treatment steps to incrementally reposition the jaw/jaws from the initial tooth arrangement to the target tooth arrangement.

METHOD FOR TRAINING LIGAND INFORMATION GENERATION MODEL, ELECTRONIC DEVICE, AND STORAGE MEDIUM

Resumen de: US20260252884A1

A method for training a ligand information generation model performed by an electronic device includes obtaining sample receptor information and sample ligand information, a binding affinity between a ligand described by the sample ligand information and a receptor described by the sample receptor information being not less than a set affinity; denoising reference noise data based on the sample receptor information using a neural network model undergoing training to obtain predicted ligand information; determining a first loss for characterizing a difference between the sample ligand information and the predicted ligand information; and training the neural network model based on the first loss to obtain a ligand information generation model, the ligand information generation model being configured to generate reference ligand information based on reference receptor information.

SYSTEMS AND METHODS FOR GENERATING MULTIMODAL DATA USING A SINGLE-TOWER ARCHITECTURE WITH A DATA GENERATION SUBSYSTEM

Resumen de: US20260253266A1

0000 A computer-implemented method of generating multimodal data. The method comprises using a token generation neural network to generate, autoregressively, an output sequence of multimodal tokens, and in response to a next multimodal token being a start-of-image token, generating an image using an image generation subsystem conditioned on features representing the current sequence of multimodal tokens obtained from the token generation neural network. The method further comprises processing the image to convert pixels of the image into a sequence of image tokens, each image token comprising a block encoding of values of the pixels in a different region of the image that maps a set of values of the pixels to a respective image token, and appending the sequence of image tokens to the current output sequence of multimodal tokens as the next multimodal tokens in the output sequence of multimodal tokens.

NEURAL NETWORK-BASED OBJECT DETECTION

Resumen de: US20260253400A1

0000 Apparatuses, systems, and techniques are presented to detect one or more objects in one or more images. In at least one embodiment, one or more neural networks can be trained to detect one or more objects, in one or more unlabeled images, based at least in part upon one or more predicted segmentations of the one or more objects.

IDENTIFYING ARTIFACTS DIFFERENCES USING GRAPH NEURAL NETWORKS

Resumen de: US20260252589A1

0000 Certain aspects of the disclosure provide for a difference analysis method. In certain aspects, a difference analysis method may include embedding a set of source documents into a knowledge graph, wherein each source document is embedded in the knowledge graph as a set of segments and a set of associations connecting two or more segments. A difference may be determined between a first segment in the set of segments of a first source document and a second segment in the set of segments of a second source document. In response to determining the difference between the first segment in the set of segments of the first source document and the second segment in the set of segments of the second source document, determining a significance of the difference on the second source document based on one or more associations of the set of associations connected to the second segment.

SAFE AND EFFICIENT MOTION PLANNING FOR AUTONOMOUS SYSTEMS USING HAMILTON-JACOBI REACHABILITY

Resumen de: US20260249882A1

0000 A method and system are provided for safe motion planning of an autonomous system using a neural-network-based approximation of a Hamilton-Jacobi (HJ) value function. The method includes computing signed distance fields (SDFs) from occupancy grid maps, deriving temporal differences between SDFs, and using these differences as input to a hypernetwork that generates parameters for a main network. The main network computes a residual of the HJ value function, which is modified using a leaky rectified linear unit and combined with a selected SDF to form an intermediate value function. A state-dependent slack function is added to produce a final HJ value function, which is used as a safety constraint in motion planning. The system enables real-time, adaptive, and robust planning in dynamic and partially observable environments.

Object-Centric Learning with Slot Attention

Resumen de: US20260252870A1

A method involves receiving a perceptual representation including a plurality of feature vectors, and initializing a plurality of slot vectors represented by a neural network memory unit. Each respective slot vector is configured to represent a corresponding entity in the perceptual representation. The method also involves determining an attention matrix based on a product of the plurality of feature vectors transformed by a key function and the plurality of slot vectors transformed by a query function. Each respective value of a plurality of values along each respective dimension of the attention matrix is normalized with respect to the plurality of values. The method additionally involves determining an update matrix based on the plurality of feature vectors transformed by a value function and the attention matrix, and updating the plurality of slot vectors based on the update matrix by way of the neural network memory unit.

GENERATING FRAMES FOR NEURAL SIMULATION USING ONE OR MORE NEURAL NETWORKS

Resumen de: US20260252876A1

Apparatuses, systems, and techniques to use one or more neural networks to generate one or more images based, at least in part, on one or more spatially-independent features within the one or more images. In at least one embodiment, the one or more neural networks determine spatially-independent information and spatially-dependent information of the one or more images and process the spatially-independent information and the spatially-dependent information to generate the one or more spatially-independent features and one or more spatially-dependent features within the one or more images.

LEARNING APPARATUS

Resumen de: US20260252891A1

A forward propagation apparatus is a forward propagation apparatus for a neural network, including: a mask generation unit that generates a binary mask; and a layer execution unit that performs an operation for a sparse convolutional layer according to a value at each coordinate of the binary mask, in which the mask generation unit: generates heat maps by performing an operation for a convolutional layer on an input feature map; generates a composite heat map obtained by combining the heat maps, into one heat map by summing up values of heat maps on a coordinate-by-coordinate basis; and generates the binary mask by binarizing a value at each coordinate of the composite heat map by using a predetermined threshold.

MULTISCALE DIMENSIONAL REDUCTION OF DATA

Resumen de: US20260252886A1

In some embodiments, a method includes segmenting updates associated with a record into a set of update subsets and generating first and second vectors based on first and second update subsets using a first neural network. The first update subset is associated with a first session and a timestamp, and the second update subset is associated with a second session. The method includes determining a first output using a second neural network based on the first and second vectors and a time difference between the first and second sessions. The method includes selecting a segment of a periodic time interval based on the timestamp, determining a second output using a third neural network based on a ratio based on the segment and the periodic time interval, and generating a characterizing vector using a fourth neural network based on the first and second outputs.

MODEL TRAINING METHOD, WATERMARK TEXT RECOGNITION METHOD, AND RELATED DEVICE

Resumen de: US20260253160A1

0000 Provided in the present application are a model training method, a watermark text recognition method, and a related device. The training method comprises: acquiring watermark style information and background style information, wherein the watermark style information is used for indicating a content style of a visible-watermark character, and the background style information is used for indicating a content style of a background image; generating a watermark image set according to a combination of the watermark style information and the background style information, wherein the watermark image set comprises a plurality of images with visible watermarks; pixelating the watermark images in the watermark image set, extracting pixel values in pixel blocks as training samples, using visible watermarks, which correspond to the watermark images, as sample labels, and combining the training samples with sample labels corresponding thereto, so as to generate a training data set; and constructing a bidirectional recurrent neural network model, and calling the training data set to train the bidirectional recurrent neural network model, so as to obtain a model, which meets a training termination condition, as a watermark restoration model, wherein the watermark restoration model is used for restoring visible-watermark characters in the images.

SPATIAL INFORMATION BASED ANOMALY DETECTION

Resumen de: US20260253200A1

0000 A method for region of interest (ROI) defect detection related to an evaluated manufactured item (MI), the method includes obtaining a reference MI image; obtaining a reference ROI definition; obtaining the evaluated MI image; feeding the reference MI image and the evaluated MI image to a neural network; detecting, by the neural network, one or more geometrical warping operations that once applied on the reference MI image results in an approximation of the evaluated MI image; applying the one or more geometrical warping operations on the reference ROI definition to provide a definition of an evaluated MI image ROI; and applying an ROI-based defect detection process on the evaluated MI image, based on the evaluated MI image ROI.

Systems and Methods for Decoding Speech from Neural Activity

Resumen de: US20260253589A1

Systems and methods for decoding speech from neural activity in accordance with embodiments of the invention are illustrated. One embodiment includes a brain-computer interface for decoding intended speech including a microelectrode array, a processor communicatively coupled to the microelectrode array, and a memory, the memory containing a speech decoding application that configures the processor to: receive neural signals from a user's brain recorded by a microelectrode array, where the neural signals comprise action potential spikes, bin the received action potential spikes by time, provide the bins to a recurrent neural network (RNN) to receive a likely phoneme at the time of each provided bin, generate an estimated intended speech using a phoneme decoder provided with the likely phonemes, where the phoneme decoder comprises a language model formatted as a weighted finite-state transducer, and vocalize the estimated intended speech using a loudspeaker communicatively coupled to the brain-computer interface.

DISTRIBUTED WEIGHT UPDATE FOR BACKPROPAGATION OF A NEURAL NETWORK

Resumen de: US20260252889A1

0000 Speed of training a neural network is improved by updating the weights of the neural network in parallel. In at least one embodiment, after back propagation, gradients are distributed to a plurality of processors, each of which calculate a portion of the updated weights of the neural network.

REINFORCEMENT LEARNING USING DENSITY ESTIMATION WITH ONLINE CLUSTERING FOR EXPLORATION

Resumen de: US20260252897A1

0000 Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network used to select actions to be performed by an agent interacting with an environment. Implementations of the described techniques can learn to explore the environment efficiently by storing and updating state embedding cluster centers based on observations characterizing states of the environment.

Self-Explaining Decision Architecture and Methods of Use

Resumen de: US20260252951A1

A Self-Explaining Decision Architecture (SEDA) for machine learning-based decision-making systems capable of generating intuitive explanations for its decisions in real time. SEDA makes use of a feature extraction subsystem and a sequence interpretation subsystem to identify patterns in data followed by a decision generation subsystem that determines appropriate actions based on those patterns. Internal state information from each of these subsystems is used to generate explanations of the system's decisions. Using this information to create explanations provides insight as to the data elements the system focused on when making decisions as well as the reasoning that was used. In at least one embodiment the system uses deep learning components including a combined convolutional neural network and long short-term memory network with attention mechanisms.

PATCHWISE MULTI-SCALE SIMILARITY ANALYSIS FOR BUILT-ENVIRONMENT STRUCTURE IMAGE LOOKUP ACROSS VARIOUS GEOGRAPHIC SCALES

Resumen de: US20260253376A1

A system and method for analyzing images of built-environment structures at multiple geographic scales using patchwise segmentation and domain-specific embedding. The system extracts image patches at a plurality of spatial scales from imagery, generates patch embeddings via a neural network trained using semi-supervised or unsupervised learning on built-environment structure images, receives a query comprising an example image or textual descriptor, computes similarity scores between query and patch embeddings using a configurable metric, and aggregates scores across scales to produce a composite similarity result. The system supports configurable weighting of scale contributions, overlapping patches for boundary fidelity, hierarchical resolution processing for bandwidth efficiency, feedback-driven model refinement, and deployment across cloud, edge, and hybrid configurations. Applications include material identification, condition assessment, damage detection, and regional trend analysis for roofing and other built-environment structures.

VIEW TRANSITION THROUGH NEURAL IMPLICIT MORPHING

Resumen de: US20260252896A1

0000 Matching keypoint pairs are generated and identified in original and rectified image spaces between two input images. A teacher neural network is trained based at least in part on the pairs of matching keypoints in the rectified image space. A neural implicit morphing network is trained jointly with the teacher neural network based at least in part on the matching keypoint pairs in the original image spaces in which predictions outputted from the teacher neural network are used to compute a loss function designated to train the neural implicit morphing network. The neural implicit morphing network on its own, after training, is caused to output intermediate images in the view transition between the two input images.

SYSTEMS AND METHODS FOR GENERATING GRAPH DATA STRUCTURES HAVING DIRECTED AND UNDIRECTED EDGES BASED ON DATASETS

Resumen de: US20260252841A1

0000 A method includes providing a semantic vector data as input to a first graph neural network to produce first prediction data for a first time, the first graph neural network including a graph data structure that has (1) a directed edge having a correlation weight and (2) an undirected edge having a causal weight, and the first graph neural network being configured to generate a first aggregation value based on a plurality of weight values associated with a plurality of nodes of the graph data structure. The semantic vector data is provided as input to a second graph neural network to produce second prediction data for a second time, the second graph neural network being produced based on the graph data structure and configured to generate a second aggregation value based on (1) the plurality of weight values and (2) a temporal dependency.

SELF-CLASSIFICATION OF NEURAL NETWORKS

Resumen de: US20260253405A1

0000 A method of classifying objects detected by n (n=2,3…) artificial neural networks in at least one image.

DYNAMICALLY PREDICTING SHOT TYPE USING A PERSONALIZED DEEP NEURAL NETWORK

Nº publicación: EP4796194A2 26/08/2026

Solicitante:

STATS LLC [US]
STATS LLC

Resumen de: EP4796194A2

A computing system retrieves ball-by-ball data for a plurality of sporting events. The computing system generates a trained neural network based on ball-by-ball data supplemented with ball-by-ball data with ball-by-ball match context features and personalized embeddings based on a batsman and a bowler for each delivery. The computing system receives a target batsman and a target bowler for a pitch to be delivered in a target event. The computing system identifies target ball-by-ball data for a window of pitches preceding the to be delivered pitch. The computing system retrieves historical ball-by-ball data for each of the target batsman and the target bowler. The computing system generates personalized embeddings for both the target batsman and the target bowler based on the historical ball-by-ball data. The computing system predicts a shot type for the pitch to be delivered based on the target ball-by-ball data and the personalized embeddings.