公开(公告)号：US12236206B1

公开(公告)日：2025-02-25

申请号：US17331478

申请日：2021-05-26

Applicant: Meta Platforms, Inc.

Inventor： Michael William Lewis ,  Marjan Ghazvini Nejad ,  Gargi Ghosh ,  Armen Aghajanyan ,  Sida Wang ,  Luke Zettlemoyer

IPC: G06F40/58 ,  G06F18/22 ,  G06N3/048 ,  G06N3/08

Abstract: In one embodiment, a method includes accessing a first document, accessing a plurality of second documents, calculating a relevance score for each of the plurality of second documents indicating a degree of relevance of the second document to the first document using an encoder of a machine-learning model, selecting a subset of the second documents based on their corresponding relevance scores, generating a target document by using the machine-learning model to process the subset of second documents and their corresponding relevance scores, and updating parameters of the machine-learning model based on a comparison between the first document and the generated target document.

公开(公告)号：US12229680B2

公开(公告)日：2025-02-18

申请号：US17035005

申请日：2020-09-28

Applicant: International Business Machines Corporation

Inventor： HsinYu Tsai ,  Stefano Ambrogio ,  Sanjay Kariyappa ,  Mathieu Gallot

IPC: G06N3/04 ,  G06N3/048 ,  G06N3/063 ,  G06N3/065 ,  G06N3/08 ,  G06N3/084

Abstract: A method comprises receiving an input signal for processing in one or more neurons of a neural network, wherein the neural network has zero bias neurons and includes a plurality of resistive processing unit (RPU) weights and each neuron has an activation function. The method also includes applying an arbitrary amplification factor to activation function outputs of the one or more neurons in the neural network, wherein the arbitrary amplification factor is based on a dynamic range of components in the neural network and compensates for conductance drift in values of the RPU weights. The method also includes performing a calculation with the neural network using the amplified activation function outputs of the one or more neurons.

公开(公告)号：US12217841B1

公开(公告)日：2025-02-04

申请号：US18150749

申请日：2023-01-05

Applicant: Brain Trust Innovations I, LLC

Inventor： David LaBorde

IPC: H04W4/02 ,  G01C21/34 ,  G05D1/00 ,  G06F17/11 ,  G06K7/10 ,  G06N3/04 ,  G06N3/048 ,  G06N3/08 ,  G06N3/088 ,  G16H10/65 ,  G16H40/63 ,  G06N3/006 ,  G06N3/044 ,  G06N3/047 ,  G06N3/084 ,  G06N3/086 ,  G06N3/126 ,  G06N5/01 ,  G06N5/048 ,  G06N7/01 ,  G06N20/10

Abstract: A system includes a plurality of tracking devices, such as RFID tags, affixed to items, such as vehicles, a data collection engine, client devices and backend devices. The backend devices include trained machine learning models, business logic, and attributes of a plurality of events. A plurality of data collection engines and systems send attributes of new events to the backend devices. The backend devices can track the items and predict particular outcomes of new events based upon the attributes of the new events utilizing the trained machine learning models.

公开(公告)号：US12217838B2

公开(公告)日：2025-02-04

申请号：US17440219

申请日：2020-12-25

Applicant: BOE TECHNOLOGY GROUP CO., LTD.

Inventor： Zhenzhong Zhang

IPC: G06F40/279 ,  G06F40/205 ,  G06N3/048 ,  G16H10/60 ,  G16H40/20

Abstract: There are provided a method and an apparatus for distributing physical examination information, an electronic device, a computer-readable storage medium, and a computer program product. The method includes: obtaining physical examination information and information of a plurality of distribution objects; inputting the physical examination information and the information of the plurality of distribution objects into an information matching model obtained by pre-training to obtain a matching degree between the physical examination information and the plurality of distribution objects; and determining a target distribution object from the plurality of distribution objects according to the matching degree between the physical examination information and each of the plurality of distribution objects, and distributing the physical examination information to the target distribution object.

公开(公告)号：US12210966B2

公开(公告)日：2025-01-28

申请号：US17035203

申请日：2020-09-28

Applicant: Robert Bosch GmbH

Inventor： Fatemeh Sheikholeslami ,  Jeremy Kolter ,  Ali Lotfi Rezaabad

IPC: G06N7/01 ,  G06N3/04 ,  G06N3/048 ,  G06N3/08

Abstract: A computer-implemented method for training a machine-learning network includes receiving an input data from a sensor, wherein the input data includes a perturbation, wherein the input data is indicative of image, radar, sonar, or sound information, obtain a worst-case bound on a classification error and loss for perturbed versions of the input data, utilizing at least bounding of one or more hidden layer values, in response to the input data, train a classifier, wherein the classifier includes a plurality of classes, including an additional abstain class, wherein the abstain class is determined in response to at least bounding the input data, outputting a classification in response to the input data, and output a trained classifier configured to detect the additional abstain class in response to the input data classifier with a plurality of classes, including an additional abstain class.

公开(公告)号：US12198460B2

公开(公告)日：2025-01-14

申请号：US17635792

申请日：2019-09-16

Applicant: INTEL CORPORATION

Inventor： Lidan Zhang ,  Qi She ,  Ping Guo

IPC: G06V40/10 ,  G06N3/044 ,  G06N3/048 ,  G06T7/12 ,  G06T7/246 ,  G06T7/73 ,  G06V10/82 ,  G06Q50/00

Abstract: Systems, methods, apparatuses, and computer program products to provide stochastic trajectory prediction using social graph networks. An operation may comprise determining a first feature vector describing destination features of a first person depicted in an image, generating a directed graph for the image based on all people depicted in the image, determining, for the first person, a second feature vector based on the directed graph and the destination features, sampling a value of a latent variable from a learned prior distribution, the latent variable to correspond to a first time interval, and generating, based on the sampled value and the feature vectors by a hierarchical long short-term memory (LSTM) executing on a processor, an output vector comprising a direction of movement and a speed of the direction of movement of the first person at a second time interval, subsequent to the first time interval.

公开(公告)号：US20250005339A1

公开(公告)日：2025-01-02

申请号：US18829480

申请日：2024-09-10

Applicant: Microsoft Technology Licensing, LLC

Inventor： Jinyu LI ,  Liang LU ,  Changliang LIU ,  Yifan GONG

IPC: G06N3/048 ,  G06F18/21 ,  G06N3/08 ,  G06N20/00 ,  G10L15/06 ,  G10L15/16

Abstract: Representative embodiments disclose machine learning classifiers used in scenarios such as speech recognition, image captioning, machine translation, or other sequence-to-sequence embodiments. The machine learning classifiers have a plurality of time layers, each layer having a time processing block and a depth processing block. The time processing block is a recurrent neural network such as a Long Short Term Memory (LSTM) network. The depth processing blocks can be an LSTM network, a gated Deep Neural Network (DNN) or a maxout DNN. The depth processing blocks account for the hidden states of each time layer and uses summarized layer information for final input signal feature classification. An attention layer can also be used between the top depth processing block and the output layer.

公开(公告)号：US12175757B2

公开(公告)日：2024-12-24

申请号：US17951971

申请日：2022-09-23

Applicant: GUANGDONG OPPO MOBILE TELECOMMUNICATIONS CORP., LTD.

Inventor： Jenhao Hsiao

IPC: G06N3/045 ,  G06N3/048 ,  G06N3/088 ,  G06V10/44 ,  G06V10/82 ,  G06V20/40 ,  G06V20/70

Abstract: A method for action localization is disclosed. The method includes dividing a video comprising a first number of frames into a second number of clips, each of the clips comprising at least one of the frames; processing the clips to obtain a first clip descriptor for the each of the clips and feature maps for each of the frames; obtaining a representation of the video based on the first clip descriptor for the each of the clips; predicting an action classification of the video based on the representation of the video; calculating an importance weight for each of the feature maps based on a gradient of the action classification; and obtaining a localization map for the each of the frames based on importance weights of corresponding feature maps.

公开(公告)号：US12175353B2

公开(公告)日：2024-12-24

申请号：US17726506

申请日：2022-04-21

Applicant: Samsung Electronics Co., Ltd.

Inventor： Hikmet Yildiz ,  Homayoon Hatami ,  Jung Hyun Bae

IPC: G06N3/0455 ,  G06N3/0464 ,  G06N3/045 ,  G06N3/048 ,  G06N3/082 ,  H03M13/27

Abstract: A symmetric interleaver for a Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) encoder and a circular padding mode are disclosed. The interleaver interleaves elements of an input block to form an output block in which an output neighborhood of elements for each element of the output block is symmetric to an input neighborhood of elements for each element of the input block. A position of an element of the input block is interleaved based on an index i of the position times a parameter δ modulo K in which the parameter δ is relatively prime with K. A test loss function may be used to train the encoder that includes a Binary Cross Entropy (BCE) loss function plus a function that minimizes a number of codeword pairs based on a Euclidean distance. The RNN encoder may be implemented as part of a Turbo Autoencoder (TurboAE) encoder.

公开(公告)号：US20240419973A1

公开(公告)日：2024-12-19

申请号：US18742948

申请日：2024-06-13

Applicant: Rain Neuromorphics Inc.

Inventor： Mohammed Elneanaei Abdelmoneem Fouda

IPC: G06N3/084 ,  G06N3/048

Abstract: A method is described. The method includes profiling a model for a learning network having a plurality of layers and associated memory. The layers include weight layers and activation layers. The plurality of weight layers including weights. The method also includes determining, based on the profiling, a training technique for the model on the learning network. The determination of the training technique includes optimizing a latency for at least one training iteration for a capacity of the associated memory. The training iteration(s) include at least one update of the weights for the weight layers.