Alerta

AUGMENTING DEEP NEURAL NETWORKS WITH RESIDUAL MEMORIZATION

Resumen de: US20260221132A1

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for performing a machine learning task. In particular, the machine learning task is performed by augmenting a trained neural network with residual memorization.

SOMATIC VARIANT PREDICTION

Resumen de: US20260221225A1

0000 Method and systems for prediction of somatic mutations based on data of a potentially tumorous sample, by receiving nucleic acid sequencing data of the potentially tumorous sample; transforming the nucleic acid sequencing data into an image-like representation; processing the image-like representation by a trained neural network to predict somatic mutations in the nucleic acid sequencing data.

SYSTEMS AND METHODS FOR INTERFACING LIVING BIOLOGICAL NEURAL NETWORKS

Resumen de: US20260220445A1

0000 Disclosed herein are systems and methods for closed-loop drug and/or environmental evaluation. In various aspects, described herein is a neuro-interface platform comprising: a neurophysiological unit comprising a multielectrode array (MEA) disposed in a chamber with biological or synthetic tissue comprising neurons. A first electrode is configured to detect an efferent signal from the neurons at a recording region corresponding to neuronal activity, and a second electrode is configured to provide electrical stimulation to the neurons at a stimulation region. The platform further includes a sensorimotor unit comprising a test device, the test device being configured to be selectively controlled based on a measurement of the efferent signal and to concurrently sense an afferent signal from a sensor coupled to the test device. The platform also includes an interface unit configured to electrically couple the neurophysiological unit and the sensorimotor unit.

SYSTEM AND METHOD FOR ESTIMATING THE METASTATIC STATUS OF A SENTINEL LYMPH NODE

Resumen de: US20260215754A1

A computer-implemented method for estimating the metastatic status of a sentinel lymph node. The method comprises: acquiring an ultrasound image of the primary tumor; segmenting said ultrasound image, wherein segmenting said ultrasound image comprises determining a region of interest, ROI, comprising an intratumoral region of the primary tumor and a peritumoral region of the primary tumor and wherein segmenting said ultrasound image comprises performing a semantic segmentation process using a CNN that was trained to determine said intratumoral region of the primary tumor; extracting a plurality of characteristics of said image from said ROI using a pre-trained convolutional neural network, CNN; selecting one or more characteristics of said plurality of characteristics; and classifying the metastatic status of said sentinel lymph node based on said one or more characteristics.

METHOD AND APPARATUS FOR PROCESSING CT IMAGE AND METHOD AND APPARATUS FOR INSPECTING INTERNATIONAL EXPRESS DELIVERY

Resumen de: US20260219417A1

0000 Provided are a method and an apparatus for processing a CT image and a method and an apparatus for inspecting the international express delivery. The method for processing the CT image includes a pre-processing step where uniform sampling and coordinate normalization are performed on three-dimensional data of the CT image to acquire pre-processed data, an object instance acquisition step where object instance data acquired after instance segmentation combined with semantic information is acquired for the pre-processed data, and a structured feature acquisition step where structured features of the object instance data are acquired using a feature extractor trained based on a neural network.

ACCELERATING MULTI-VIEW SEARCH WITH DYNAMIC VIEW ROUTING

Resumen de: US20260220430A1

Methods, systems, and apparatuses, including computer programs encoded on computer storage media, for selecting relevant documents. A system processes a query using a query encoder neural network to generate query embeddings. A router neural network processes the query embeddings to generate a respective score for each embedding. The system selects a representative query embedding based on the scores and utilizes the representative query embedding and stored document embeddings to generate a relevance score for each of a plurality of documents. The system then selects a subset of the documents as relevant to the query based on the relevance scores.

DILATOR PROXIMAL END BLOCKER IN A SURGICAL VISUALIZATION

Resumen de: AU2024406878A1

Computing systems and methods are disclosed for detecting and tracking a proximal end of a dilator from a surgical video stream, and mitigating the tool's visually distracting effects. In one example system, a memory may store instructions that, when executed by one or more processors, may cause the system to receive, in real-time, image data from an incoming surgical video stream of a field of view of the digital surgical microscope camera. The field of view may show a proximal end of a dilator. The system may generate, using the image data and a trained neural network, a bounding box around a region corresponding to the proximal end of the dilator. A pixel modification algorithm may be applied to the region of the image data corresponding to the proximal end of the dilator. An updated image data may be generated in real-time for an outgoing surgical video stream.

Proxy-Based Multi-Modal Learning For Render Time Prediction

Resumen de: US20260220485A1

0000 Techniques for predicting render times are disclosed. A system accesses auxiliary rendered outputs that were generated from a coarse render pass of a 3D scene description using a set of hardware resources. The system accesses system performance metrics of the set of hardware resources associated with the coarse render pass. The system encodes visual feature representations from the auxiliary rendered outputs using a convolutional neural network. The system also encodes the system performance metrics and a set of rendering configuration parameters associated with a target rendering operation. The system generates, using a multi-modal fusion model, a fused representation of the encoded visual feature representations, the encoded system performance metrics, and the encoded first set of rendering configuration parameters, using an attention-based mechanism that weights contributions of the inputs to the model. The system predicts a render time for the target rendering operation based on the fused representation.

TITER PROCESSING METHOD AND SYSTEM BASED ON DEEP LEARNING

Resumen de: US20260220770A1

0000 The present application discloses a titer processing method and system based on deep learning, which belong to the field of machine learning, and comprises: obtaining a first image of non-syphilis treponemal serological test results; judging the clarity of the test image and outputting a second image; applying a convolutional neural network on the second image for detecting circular targets and outputting the center coordinates and radius of the circular targets; segmenting the second image according to the center coordinates and radius of the detected circular targets to extract circular target regions; sorting the extracted circular target regions according to preset sorting rules; using a convolutional neural network to classify the sorted circular target regions and outputting a negative or positive result; calculating a corresponding titer of the negative or positive result.

NEURAL NETWORK TRAINING METHOD AND DEFECT DETECTION METHOD AND APPARATUS

Resumen de: US20260220931A1

A method of training a neural network is provided. The method includes steps of obtaining labeled defect information of an object and a training image set collected from the object, and obtaining, based on the training image set, feature image sets for representing a plurality of features of the object; inputting the feature image sets into the neural network, utilizing attention mechanism modules in the neural network to carry out local attention generation and global attention generation with respect to the feature image sets, respectively, so as to generate processing results, and creating, based on the processing results, training defect information of the object; and comparing the training defect information of the object and the labeled defect information of the object, so as to train the neural network and adjust parameters of the neural network.

OPTIMIZED PLACEMENT FOR EFFICIENCY FOR ACCELERATED DEEP LEARNING

Resumen de: US20260219967A1

0000 Techniques in optimized placement for efficiency for accelerated deep learning provide improvements in one or more of accuracy, performance, and energy efficiency. An array of processing elements comprising a portion of a neural network accelerator performs flow-based computations on wavelets of data. Each processing element comprises a compute element to execute programmed instructions using the data and a router to route the wavelets. The routing is in accordance with virtual channel specifiers of the wavelets and controlled by routing configuration information of the router. A software stack determines optimized placement based on a description of a neural network. The determined placement is used to configure the routers including usage of the respective colors. The determined placement is used to configure the compute elements including the respective programmed instructions each is configured to execute.

SYSTEMS AND METHODS FOR EFFICIENT MODEL EXECUTION ON MACHINE-LEARNING ACCELERATORS

Resumen de: US20260220439A1

Methods, systems, and apparatus for reducing latency of configuration of image sensor and image signal processor. A computing system can include a machine learning (ML) processing engine that can process denoising diffusion ML models for execution on statically compiled ML accelerators. The system can determine that a partitioned graph representation, which includes connected subgraphs that each represent at least one layer of the neural network of the ML model, forms a directed acyclic graph. The system can insert cache nodes in the graph representation, where each cache node corresponds to a respective subgraph and is configured to cache output of the respective subgraph. The system can generate an execution dataflow graph including a plurality of iterations of the partitioned graph representation, where, at one or more iterations during model inference operations, the execution dataflow graph uses inputs from cache nodes and excludes execution of subgraphs corresponding to the cache nodes.

HIERARCHICAL MEMORY ATTENTION NETWORK WITH MULTI-COMPONENT WEIGHTED ATTENTION AND ADAPTIVE RESOURCE ALLOCATION

Resumen de: US20260220425A1

Systems, methods, and computer-readable media are disclosed for providing hierarchical processing in neural network architectures (e.g., transformer-based models, recurrent networks, state space models, memory-augmented networks, and retrieval-augmented systems), which may improve computational efficiency and memory utilization.

Controlling a vehicle at an entry to a roundabout

Resumen de: GB2703381A

The invention relates to controlling a vehicle at an entry to a roundabout. The invention includes receiving 201 dynamic data of dynamic characteristics related to how a further vehicle is traversing the roundabout. This is input into a first branch of a dual-branch neural network, which determines 203 a plurality of dynamic data vectors indicative of temporal dynamics of the further vehicle. Static data characteristics associated with the roundabout is input into a second branch of the neural network, and a static data vector indicative of contextual information associated with the roundabout is determined 204. At an attention layer of the neural network, a cross-attention mechanism is applied 205 to the dynamic data vectors and the static data vector to obtain a cross-attended vector. A cross prediction indicating whether the further vehicle will cross in front of the vehicle is then determined 206, and a vehicle control signal is output 207 in dependence thereof. Fig. 2

METHOD AND SYSTEM FOR GENERATING TECHNICAL EXPLANATIONS OF ALARMS IN A SCADA SYSTEM

Resumen de: EP4783068A1

0001 According to an aspect of the present inventive concept there is provided a computer-implemented method (1000) for generating technical explanations of alarms in a supervisory control and data acquisition, SCADA, system for an electrical infrastructure, the method (1000) comprising: providing (1100) a neural network, NN, model trained to detect anomalies in data, wherein the NN model is trained on a training dataset comprising training data based on operational data retrieved by the SCADA system from the electrical infrastructure and comprising data related to a plurality of system parameters of the SCADA system, receiving (1200) run data based on operational data retrieved by the SCADA system from the electrical infrastructure, the run data comprising data related to the plurality of system parameters, detecting (1300), with the NN model, an anomaly in the run data, and generating (1400) a technical explanation of the detected anomaly by means of an interpretability model, wherein the interpretability model applies a model-agnostic interpretable technique to the run data to determine a contribution of the system parameters in the plurality of system parameters to the detection of the anomaly.

A NEURAL NETWORK SYSTEM WITH MULTIPLE INPUTS AND MULTIPLE OUTPUTS

Resumen de: CN121889838A

A method and apparatus for deep learning. A first input and a second input are accessed. A first embedding of the first input is generated using the bound network. A second embedding of the second input is generated using the bound network. The first embed and the second embed are aggregated to generate a combined embed. The transform function is applied to the combinatorial embedding to generate a transformed combinatorial embedding. The combined embedding of transforms is processed using an unbinding network to extract an embedding of a first transform for the first input and an embedding of a second transform for the second input. An inference function is applied to the embedding of the first transform to generate a first output. An inference function is applied to the embedding of the second transform to generate a second output.

SELF-CLASSIFICATION OF NEURAL NETWORKS

Resumen de: WO2025083107A1

The invention relates to a method for categorizing objects detected by n (n = 2, 3...) artificial neural networks in at least one image.

PARALLEL COMPUTING SCHEME GENERATION FOR NEURAL NETWORKS

Resumen de: EP4783073A2

0001 Various embodiments relate to determining a parallel computation scheme for a neural network. A device may receive a computation graph and transform the computation graph into a dataflow graph comprising recursive subgraphs. Each recursive subgraph may comprise a tuple of another recursive subgraph and an operator node, or an empty graph. The device may determine a number of partitioning recursions based on a number of parallel computing devices. For each partitioning recursion, the device may determine costs corresponding to operator nodes, determine a processing order of the recursive subgraphs, and process the recursive subgraphs. To process a recursive subgraph, the device may select a partitioning axis for tensor(s) associated with an operator node of the recursive subgraph. The device may output a partitioning scheme comprising partitioning axes for each tensor associated with the operator nodes. Devices, methods, and computer programs are disclosed.

FPGA-ORIENTED DSP PLACEMENT OPTIMIZATION METHOD FOR CNN ACCELERATORS

Resumen de: US20260212104A1

An FPGA-oriented DSP placement optimization method for CNN accelerators includes the following steps: S1. DSP path information extraction: converting a designed netlist into a graph representation, and carrying out data-path DSP node identification and data-path DSP graph building; and S2. datapath-driven DSP placement: distributing data-path DSP nodes to specific positions on an FPGA according to an extracted data-path DSP graph. According to the invention, automated extraction and building of data-path DSP graphs are carried out by means of graph neural network (GCN)-based DSP node classification and min-cost flow (MCF) model optimization algorithms, and compact placement and cascade constraint optimization are used in combination, thereby greatly improving the timing performance and computing efficiency of the placement and also significantly improving the clock frequency and throughput. Therefore, the method provides a universal and efficient FPGA placement solution for multiple CNN accelerator architectures.

SELF-DEVELOPING NEURAL NETWORK DESIGN METHOD INSPIRED BY DNA DAMAGE REPAIR MECHANISM, AND STORAGE MEDIUM

Resumen de: WO2026152352A1

A self-developing neural network design method inspired by a DNA damage repair mechanism, and a storage medium. The method comprises: selecting a neural network model and defining hyperparameters thereof; acquiring an image dataset, and dividing the image dataset into a training set, a validation set and a test set, in order to form sequential tasks; on the basis of data of the first task in the training set, training the neural network model; on the basis of data of the first task in the validation set, testing a trained neural network model; performing self-developing growth on the neural network model in the width and depth directions, in order to increase the scale of the neural network model; repeating the above training and testing processes, determining whether it is necessary to further increase the scale of the neural network model, and if the increase is stopped, acquiring the classification accuracy of the model based on task data in the test set; and repeating the above process until the testing of all the task data in the test set is completed. The scale of a neural network model is dynamically adjusted during a training process, and new neurons are grown to adapt to changing tasks and environments, thereby promoting the intelligent development of neural networks in image classification tasks and the practical application thereof.

METHODS AND APPARATUSES FOR CHARACTERIZING CHEMICAL SUBSTANCES, MEASURING PHYSICOCHEMICAL PROPERTIES AND GENERATING CONTROL DATA FOR SYNTHESIZING CHEMCIAL SUBSTANCES

Resumen de: US20260212962A1

Provided is a method for generating a digital representation of a chemical substance. This may involve training aspects of neural networks to represent chemical sub-stances. Provided further relates to applications of the digital representation including a computer program product, a database search engine for identifying chemical sub-stances, apparatuses for generating measurement data associated with chemical substances and control data associated with synthesis specifications for chemical substances.

AUTOMATIZED DETECTION OF INTESTINAL INFLAMMATION IN CROHN'S DISEASE USING CONVOLUTIONAL NEURAL NETWORK

Resumen de: US20260212495A1

0000 The invention relates to system and methods for predicting and/or diagnosing of IBD from ultrasound images according to one or more of diagnostic signs.

METHOD AND APPARATUS FOR COMPRESSING A NEURAL NETWORK, ELECTRONIC DEVICE, AND AIRCRAFT

Resumen de: US20260212183A1

A method and an apparatus for compressing a neural network. Circuitry receives a neural network comprising a set of parameters being in at least one floating-point number format of a first precision. The circuitry also applies at least one mathematical model on the neural network to determine errors introduced by using arithmetic of a second precision lower than the first precision and propagation through network layers of the neural network due to using the arithmetic of the second precision instead of the first precision. The circuitry then applies a solver to an optimization problem formulated based on the errors determined by the at least one mathematical model to determine a set of optimized parameters. Finally, the circuitry compresses the neural network based on the set of optimized parameters to output a compressed neural network.

METHOD AND SYSTEM OF IMAGE CLASSIFICATION USING A DEEP NEURAL NETWORK EMBEDDED WITH MULTI-SCALE SPATIAL ATTENTION MECHANISM

Resumen de: US20260212659A1

0000 The present invention provides a method for image classification by incorporating a deep neural network embedded with multiscale spatial attention mechanism (MSSAM). The method according to the present invention comprises various stages: Stage I—Data preparation stage; Stage II—Model training stage; Stage III—Evaluation and Testing stage; Stage IV—Iterative optimization stage. During Data preparation stage, data is collected from a large and diverse dataset of images relevant to specific classification task. During the Model training stage, the model architecture is established, appropriate loss function is selected, an optimizer and an initial learning rate is chosen, the model is trained on training dataset, monitoring validation performance and experiments are performed with hyperparameters. During the Evaluation and Testing stage, model's performance is evaluated on the validation set using metrics and during Iterative optimization stage, the optimization process is iterated and continuously monitored for best results.

MODULAR CONCEPT-BASED LANGUAGE PROCESSING WITH TRACE-DRIVEN RESOURCE TIERING AND DYNAMIC CACHING

Nº publicación: US20260211817A1 23/07/2026

Solicitante:

RIVKIN LEON [US]
Rivkin Leon

Resumen de: US20260211817A1

0000 Modular systems and methods for concept-based processing of natural language are provided. An ontology dictionary stores concepts each having a concept identifier, attributes, and typed relationships. Natural language inputs are mapped to concept identifiers, optionally by deriving intermediate units using transforms or statistical analysis and mapping the intermediate units to concepts, or by mapping raw tokens to concepts. A neural network model processes the concept identifiers to generate outputs. During inference and/or training, a trace engine records activation information associated with concepts and stores activation information or derived statistics in a trace data structure. A resource controller uses the trace data structure to reconfigure, during inference for subsequent inputs, a resource allocation policy that allocates frequently used concept data to a faster memory tier and allocates infrequently used concept data to a slower tier or specialized tail handling. Predictive prefetching and relational group caching may be performed.