Alerta

METHOD FOR SECURING A COMPUTER SYSTEM AGAINST INTEGRITY ATTACKS

Absstract of: EP4811192A1

0001 The present invention relates to a method for securing against adversarial and backdoor attacks a computer system including a memory storing a plurality of benign images (501) and using at least one neural network comprising an input layer, hidden layers, and an output layer generating an embedding or a classification result into a set of output classes for an input image presented at the input layer, said method being performed by said computer system and comprising, before or during the inference phase: - splitting at least one benign image (501) among said stored benign images into a plurality of zones according to at least one predefined method; and at the inference phase: - acquiring an image (502); - splitting the acquired image into a plurality of zones according to said at least one predefined method; -generating at least one composite image (503, 504, 505, 506) by replacing, in at least one split benign image among said stored benign images a zone of the split benign image corresponding to a zone of the split acquired image with said zone of the split acquired image; -applying said at least one neural network to said at least one generated composite image for generating a result from said output layer of said at least one neural network; -comparing said generated result and a result generated by applying said at least one neural network to said acquired image; - based on said comparison, detecting an adversarial or backdoor attack and performing a predetermi

GENERATING TRAINING DATA FOR ESTIMATING MATERIAL PROPERTY PARAMETER OF FABRIC AND ESTIMATING MATERIAL PROPERTY PARAMETER OF FABRIC

Absstract of: EP4811314A2

Estimating a material property parameter of fabric involves receiving information including a three-dimensional (3D) contour shape of fabric placed over a 3D geometric object, estimating a material property parameter of the fabric used for representing drape shapes of 3D clothes made by the fabric by applying the information to a trained artificial neural network, and providing the material property parameter of the fabric.

OPEN-VOCABULARY OBJECT DETECTION IN IMAGES

Absstract of: EP4811306A2

0001 Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for object detection. In one aspect, a method comprises: obtaining: (i) an image, and (ii) a set of one or more query embeddings, wherein each query embedding represents a respective category of object; processing the image and the set of query embeddings using an object detection neural network to generate object detection data for the image, comprising: processing the image using an image encoding subnetwork of the object detection neural network to generate a set of object embeddings; processing each object embedding using a localization subnetwork to generate localization data defining a corresponding region of the image; and processing: (i) the set of object embeddings, and (ii) the set of query embeddings, using a classification subnetwork to generate, for each object embedding, a respective classification score distribution over the set of query embeddings.

COMPUTER-IMPLEMENTED METHOD FOR APPROXIMATING AT LEAST TWO UNKNOWN VARIABLES OF A SET OF PARTIAL DIFFERENTIAL EQUATIONS, HYBRID COMPUTING SYSTEM, COMPUTER PROGRAM PRODUCT AND COMPUTER-READABLE MEDIUM

Absstract of: CA3301830A1

The invention relates to a computer-implemented method for approximating at least two unknown variables (10) of a set of partial differential equations using a quantum physics-informed neural network (100), the method comprising the steps: providing a numerical grid (20) having grid coordinates (30) at which the unknown variables (10) are to be approximated; providing a quantum physics-informed neural network (100), the quantum physics-informed neural network (100) comprising a hybrid network (110) for each unknown variable (10), wherein the hybrid networks (110) are not interconnected to each other, each hybrid network (110) having a quantum network (120) having at least one quantum layer(130) and a classical network (140) having at least one classical layer (150), wherein the respective quantum network (120) and the respective classical network (140) are not interconnected; inputting the grid coordinates (30) into the quantum physics-informed neural network (100), such that the grid coordinates (30) are input to each hybrid network (110); and computing the output (Hout) of each hybrid network (110) for each grid coordinate (30), each output (Hout) corresponding to a different unknown variable (10) to be approximated at the grid coordinate (30), wherein the output (Hout) of each hybrid network (110) is obtained by combining an output of the respective quantum network (Qout) and an output of the respective classical network (Cout) of said hybrid network (110). The invention f

LASER RADAR TARGET DETECTION METHOD, SYSTEM, TERMINAL, AND STORAGE MEDIUM

Absstract of: US20260276826A1

The method comprises: acquiring target point cloud data and determining whether the total number of points in the target point cloud data exceeds a maximum point limit; when the total number of points in the target point cloud data exceeds the maximum point limit, performing weighted sampling on the target point cloud data using a point cloud sampling method based on the dynamic fusion of density and depth weights to obtain sampled point cloud data; performing voxelization on the sampled point cloud data to obtain voxelized point cloud data, processing the voxelized point cloud data using an improved feature encoding module, and outputting a target tensor; generating a pseudo-image based on the target tensor, extracting features from the pseudo-image using a two-dimensional convolutional neural network, and performing detection and regression on the extracted features using a detection head to obtain detection results.

NEURAL NETWORK MODEL TRAINING METHOD INCORPORATING PERIODIC FEATURES AND PARTIAL NORMALIZATION, AND NEURAL NETWORK MODEL

Absstract of: WO2026189009A1

The present application provides a neural network model training method incorporating periodic features and partial normalization, and a neural network model. An input layer of the model is a self-attention layer. The method comprises: upon receiving training samples, calling a self-attention layer to extract periodic features and first aperiodic features comprised in the training samples; normalizing a plurality of feature values of the first aperiodic features to obtain second aperiodic features; decomposing the periodic features into sine wave features and cosine wave features; concatenating the plurality of second aperiodic features, the sine wave features and the cosine wave features, and using concatenated feature data as training data of a neural network model to be trained; and using the training data to train said neural network model to obtain a target neural network model. The embodiments of the present application can remove the impact of normalization on capturing periodic features, and improve the effect of periodic modeling outside the range of training data.

FUSION IMAGE GENERATING METHOD AND APPARATUS, AND STORAGE MEDIUM STORING INSTRUCTIONS FOR PERFORMING FUSION IMAGE GENERATING METHOD

Absstract of: US20260279031A1

A fusion image generating method using a fusion image generating apparatus, according to an embodiment, comprises the steps of: performing temporal and spatial synchronization of color images and thermal images and preprocessing into preset formats; generating multidimensional color image spectrum features from the preprocessed color images by using a color image encoder in a pre-trained neural network; generating a multidimensional thermal image spectrum feature from the preprocessed thermal images by using a thermal image encoder in the pre-trained neural network; and generating fusion images by fusing the multidimensional color image spectrum features and the multidimensional thermal image spectrum features by using a fusion image decoder in the pre-trained neural network.

SYSTEM AND METHOD FOR ARTIFICIAL INTELLIGENCE GOVERNANCE PLATFORM FOR RESPONSIBLE ARTIFICIAL INTELLIGENCE

Absstract of: US20260278136A1

An Artificial Intelligence Governance system, ensuring responsible AI utilization is disclosed. The system features a processing subsystem hosted on a server, orchestrating bidirectional communications across a network among numerous modules. A database module captures and manages diverse activities associated with interacting with Neural network with attention based AI models, while a user input module facilitates user engagement through API or UI. A configuration module defines organizational policies and regulatory requirements, ensuring responsible AI practices. An authentication module regulates user access based on roles, employing granular permission sets. A Neural network with attention based AI governance module uses a pre-trained AI model to classify risks in user prompts. A session management module establishes standardized API endpoints for seamless integration, and a dashboard module offers a summarized overview, governing responsible AI usage within the organization. This system provides transparency, and ethical AI practices across various organizational contexts.

METHOD FOR PROVIDING INFORMATION ABOUT IN VITRO FERTILIZED EMBRYO SCREENING, AND APPARATUS USING SAME

Absstract of: US20260279547A1

Provided are a method and an apparatus for providing information about in vitro fertilized embryo screening using embryo images. The method, executed by a processor, includes the steps of: receiving an original embryo image; extracting, from the original embryo image, both an inner cell mass (ICM) image and a trophectoderm (TE) image; converting the original embryo image, the extracted ICM image, and the extracted TE image into a binary image or gray scale images; and by using a convolutional neural network model pre-trained to screen an in vitro fertilized embryo by using converted original embryo images, converted ICM extracted images, and converted TE extracted images as inputs, determining an in vitro fertilized embryo by using the converted original embryo image, the converted ICM extracted image, and the converted TE extracted image as inputs.

UNIFIED VISUAL UNDERSTANDING AND IMAGE GENERATION NEURAL NETWORKS

Absstract of: WO2026193311A1

Systems and methods for performing both visual understanding tasks and image generation tasks using the same multi-modal neural network.

AUTONOMOUS ARTIFICIAL INTELLIGENCE (AI) AGENT CONTROL OF MULTITIER AI INFERENCE DISAGGREGATION

Absstract of: US20260278423A1

A distributed artificial-intelligence inference system includes a master autonomous AI agent configured to control disaggregation of inference across multiple execution tiers. An input is received at an edge tier, and the master autonomous AI agent generates a plurality of candidate plans or branch continuations and instantiates corresponding sub-agents. The system identifies a shared inherited state common to multiple branches and branch-specific state corresponding to suffix portions after divergence. A promotion policy selects a promoted subset of branches for continuation at a data-center tier while maintaining a non-promoted subset at the edge tier. The promotion policy may evaluate branch score, local resource cost, transfer efficiency of shared-prefix state, model compatibility between source-tier and destination-tier models, branch type, network condition, and quality-related factors. A promotion package is generated including a shared-prefix reusable inference-state portion and separate branch-specific suffix portions for respective promoted branches. The data-center tier continues execution of the promoted subset using the promotion package. In some embodiments, the data-center tier further disaggregates inference among compute-dense, memory-centric, and persistence-oriented clusters. In some embodiments, an inference processor reconfigures an on-chip memory array between scratchpad mode and cache mode according to neural-network layer type.

GENERATIVE DRIVING MODEL

Absstract of: US20260279072A1

A computing device of a vehicle may receive a sequence of images. The computing device can execute a neural network to generate a first plurality of tokens from the images. The computing device can input the tokens into a machine learning model, such as a machine learning language processing model (e.g., a large language model or a neural network including a transformer model). The machine learning model can output a second plurality of tokens based on the execution. In doing so, the machine learning model can output one or more characters or words that represent a future state of the objects in an image. The computing device can analyze the second plurality of tokens. The computing device generate an alert or control the vehicle accordingly.

MULTI-MODAL NEURAL NETWORKS WITH DECODER-ONLY LANGUAGE MODELS

Absstract of: US20260278340A1

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a multi-modal neural network using contrastive and image captioning losses.

PARTIAL DISCHARGE CLASSIFICATION METHOD, DEVICE, AND COMPUTER PROGRAM USING ARTIFICIAL NEURAL NETWORK MODEL

Absstract of: WO2026192201A1

This partial discharge classification method using an artificial neural network model comprises the steps of: generating a prescribed pattern by analyzing time-series data for an acoustic signal; and classifying the type of partial discharge corresponding to the prescribed pattern when the prescribed pattern is input to a trained artificial neural network model.

Systems and Methods for Efficient Differentiable Logic Gate Networks

Absstract of: US20260278370A1

Systems and methods are disclosed for implementing and training differentiable logic-gate neural networks. A network includes hyper-logic gate nodes configured as differentiable lookup tables that receive selector signals and LUT entry signals and generate outputs via differentiable selection, enabling LUT-style multiplexing during discretization while remaining trainable by gradient-based optimization. LUT entry signals may be learned parameters, outputs of other nodes, external inputs, or internal memory-state values, enabling aggregation, switching, skip connections, and memory read/write operations. Training may employ constrained or reduced parameterizations that transform unconstrained trainable parameters via elementwise bounding functions and coefficient-mapping matrices to produce coefficient vectors for k-input gate functions, reducing storage and computation relative to enumerating discrete gates. Mixed-precision training may compute forward activations at lower precision while performing backpropagation at higher precision with selective storage or recomputation.

ARTIFICIAL INTELLIGENCE DEVICE FOR GROUP-EQUIVARIANT QUANTUM NEURAL FUNCTIONALS FOR DENSITY FUNCTIONAL THEORY SIMULATION AND METHOD THEREOF

Absstract of: US20260278441A1

A method for controlling an artificial intelligence (AI) device can include generating a real-space electron density for a target system, encoding a representation of the real-space electron density into a multi-qubit quantum state, processing the multi-qubit quantum state using a parameterized quantum circuit, in which the parameterized quantum circuit includes a quantum neural network configured to be equivariant to an orthogonal group O(3) or finite subgroups of the orthogonal group O(3), measuring an output of the parameterized quantum circuit to obtain spatial weighting coefficients, calculating an exchange-correlation (XC) energy contribution by combining the spatial weighting coefficients with one or more of analytical primitive energy-density terms, and outputting the exchange-correlation (XC) energy contribution.

3D MULTI-SENSOR PERCEPTION AND OBJECT TRACKING

Absstract of: US20260279058A1

In various examples, the embodiments disclosed herein describe a 3D perception-based machine learning framework for generating behavior data for detected objects (e.g., objects, persons, animals, machines, etc.) using multi-view optical image data. The framework processes multi-view optical image data from multiple camera sensors, and neural network-based spatial-temporal processing, to generate object behavior data that may be used to facilitate accurate real-time multi-target multi-camera (MTMC) object tracking across a monitored environment. The framework may comprise one or more machine learning models that input multi-view image sensor data and infer behavior data that may include, but is not limited to, 3D bounding shapes, instance features, and/or 3D Re-Identification (ReID) feature embeddings that may be used for assigning an object ID and for extending tracking of detected objects within the monitored environment. To generate an ReID feature embedding, an ReID module may aggregate features from different camera views based on visibility scores.

Method for Calculating Absorption Rate of Super Absorbent Polymer on Basis of Image Classification Algorithm

Absstract of: US20260279004A1

A method and system for calculating an absorbing speed of a predetermined absorbent material from video data obtained by photographing a video of the absorbent material absorbing water. Absorption video data is separated into each frame image data, an artificial neural network model is used to determine whether each frame of image data is an absorption-in-progress image or an absorption-completion image, and an absorption start time point and an absorption end time point are detected, thereby calculating the absorption speed of the absorbent material.

SYSTEM AND METHOD FOR SEGMENTATION OF MAGNETIC RESONANCE IMAGES USING DEEP LEARNING WITH DOMAIN ADAPTATION

Absstract of: WO2026193213A1

Systems and methods disclosed herein are directed to segmentation of magnetic resonance (MR) images of a subject and include receiving at least one MR image of the subject, providing the at least one MR image of the subject to a segmentation system comprising a neural network-based unsupervised domain adaptation (UDA) model, a foundation model, and a mask- guided semi-supervised (MGSS) network, generating a segmentation of a region of interest of the at least one MR image of the subject using the segmentation system and displaying the segmentation of the at least one MR image of the subject.

APPARATUS AND METHOD FOR PROCESSING SCALABLE FEATURE MAPS USING ARTIFICIAL NEURAL NETWORKS

Absstract of: US20260279040A1

A neural processing unit (NPU) for decoding video or feature map is provided. The NPU may comprise at least one processing element (PE) to perform an inference using an artificial neural network. The at least one PE may be configured to receive and decode data included in a bitstream. The data included in the bitstream may comprise data of a base layer. Alternatively, the data included in the bitstream may comprise data of the base layer and data of at least one enhancement layer. The data of the base layer included in the bitstream may include a first feature map. The data of the at least one enhancement layer included in the bitstream may include a second feature map.

BLOCKCHAIN-BASED DEEP NEURAL NETWORK MODEL FEDERATED RECOMBINATION METHOD AND SYSTEM

Absstract of: US20260278394A1

0000 A blockchain-based deep neural network model federated recombination method includes constructing a consortium blockchain for model federated recombination; configuring a smart contract for the model federated recombination; storing data in a model federated recombination process; training a source model of a model training party, and providing the source model to a model recombination verification party for federated combination; performing model federated recombination, wherein the model recombination verification party performs federated recombination on the source model to obtain a target model; and calculating a contribution value of the model training party. For the reusability of a deep neural network model and the finiteness of computing power and data during single-node user model recombination, multi-node model federated recombination is performed. For the data security and privacy in a federated process and the mutual distrust between multiple nodes in federation, a Hyperledger Fabric-based consortium blockchain and a smart contract are constructed.

GENERATING SEGMENTATION MASKS FOR OBJECTS IN DIGITAL VIDEOS USING POSE TRACKING DATA

Absstract of: US20260279093A1

The present disclosure relates to systems, methods, and non-transitory computer-readable media that generate joint-based segmentation masks for digital objects portrayed in digital videos. In particular, in one or more embodiments, the disclosed systems utilize a video masking model having a pose tracking neural network and a segmentation neural network to generate the joint-based segmentation masks. To illustrate, in some embodiments, the disclosed systems utilize the pose tracking neural network to identify a set of joints of the digital object across the frames of the digital video. The disclosed systems further utilize the segmentation neural network to generate joint-based segmentation masks for the video frames that portray the object using the identified joints. In some cases, the segmentation neural network includes a multi-layer perceptron mixer layer for mixing visual features propagated via convolutional layers.

Wearable Electronic Device with Built-in Intelligent Monitoring Implemented using Deep Learning Accelerator and Random Access Memory

Absstract of: US20260278362A1

Systems, devices, and methods related to a deep learning accelerator and memory are described. For example, a wearable electronic device may be configured to execute instructions with matrix operands and configured with: a housing to be worn on a person; a sensor having one or more sensor elements generate measurements associated with the person; random access memory to store instructions executable by the deep learning accelerator and store matrices of an artificial neural network; a transceiver; and a controller to monitor an output of the artificial neural network, generated using the measurements as an input to the artificial neural network. Based on the output, the controller may control selective storage of measurement data from the sensor, and/or selective communication of data from the wearable electronic device to a separate computer system.

ENVIRONMENT GENERATION USING ONE OR MORE NEURAL NETWORKS

Absstract of: US20260278952A1

Apparatuses, systems, and techniques are presented to generate a simulated environment. In at least one embodiment, one or more neural networks are used to generate a simulated environment based, at least in part, on stored information associated with objects within the simulated environment.

TRAINING NEURAL NETWORKS USING LINEAR COMPANION NODES TO MITIGATE GRADIENT ATTENUATION

Nº publicación: US20260278381A1 17/09/2026

Applicant:

D5AI LLC [US]
D5AI LLC

Absstract of: US20260278381A1

A system and method for controlling a nodal network. The method includes estimating an effect on the objective caused by the existence or non-existence of a direct connection between a pair of nodes and changing a structure of the nodal network based at least in part on the estimate of the effect. A nodal network includes a strict partially ordered set, a weighted directed acyclic graph, an artificial neural network, and/or a layered feed-forward neural network.