УСТРОЙСТВО ДЛЯ ДИФФЕРЕНЦИРОВАННОГО ВНЕСЕНИЯ ФУНГИЦИДОВ И БИОПРЕПАРАТОВ В СЕЛЬСКОХОЗЯЙСТВЕННЫЕ КУЛЬТУРЫ

Полезная модель относится к сельскохозяйственному машиностроению, в частности к автоматизированным устройствам для выборочного внесения фунгицидов и биопрепаратов в сельскохозяйственные культуры, на основе технологий дифференцированного внесения препаратов, которые позволяют оптимизировать применение средств защиты растений, рассчитанные по специальным алгоритмам и методикам. Технический результат, достигаемый при использовании заявляемой полезной модели, заключается в снижении расхода препаратов с повышением точности их внесения и, как следствие, качества обработки культур. Заявленный технический результат достигается тем, что устройство для внесения фунгицидов и биопрепаратов в сельскохозяйственные культуры, включающее корпус, установленный на раме, снабженной колесами с электромоторами, блок управления, источник питания и блок опрыскивания, включает линейные актуаторы, установленные в нижней части корпуса над мотор-колесами, и снабжено средством распознавания очагов поражения сельскохозяйственных ...

Система идентификации личности по изображению лица, сформированного сверточной искусственной нейронной сетью

Изобретение относится к системе идентификации личности по изображению лица, сформированного сверточной искусственной нейронной сетью. Техническим результатом является обеспечение идентификации личности по изображению лица. Система содержит модуль приема цифрового описания изображения лица личности с информационного выхода сверточной искусственной нейронной сети, модуль селекции адресов записей идентификационных данных личностей в базе данных сервера системы, модуль памяти, модуль приема адресов считывания данных, модуль сравнения данных, модуль приема идентификационных данных из сервера базы данных, модуль выдачи документальных данных личности и сервер базы данных. 4 ил.

Способ диагностики пронации пяточной кости

Изобретение относится к медицинской технике, а именно к способам ортопедической диагностики пронации пяточных костей стоп. Способ диагностики пронации пяточных костей у детей, в котором проводят клинический осмотр нижних конечностей. Далее делают фотографию дорсальной поверхности нижних конечностей в положении стоя на горизонтальной поверхности. Затем фотоизображение обрабатывают с помощью входящей в библиотеку PyTorch Keypoint RCNN модели машинного обучения путем выделения ключевых точек и линий дорсальной поверхности голеностопного сустава и стопы. Далее находят координаты точек, используя Keypoint RCNN модель, а обучение данной модели произведено с помощью машинного обучения. Данная модель натренирована на поиск 17 ключевых точек дорсальной поверхности нижних конечностей, а обучение произведено с помощью PyTorch и Scikitleam. Затем анализируют и рассчитывают углы пронации пяточных костей. Техническим результатом является доступность и повышение диагностической точности определения угла ...

СПОСОБ ПОСТОБРАБОТКИ СЕЙСМИЧЕСКИХ ДАННЫХ

Изобретение относится к области геофизических исследований и, в частности, к сейсмической интерпретации, является технологией постобработки сейсмических данных с использованием методов машинного обучения для улучшения их качества и повышения точности структурной интерпретации геологических структур. Изобретение может быть использовано для целей разведочной геофизики. Согласно изобретению осуществляют генерацию исходных синтетических сейсмических данных с широким диапазоном частот вейвлета, используемого для генерации. Затем добавляют в исходные синтетические сейсмические данные различные типы шумов, артефактов обработки структурных и тектонических объектов. Далее осуществляют обучение конволюционной нейронной сети на датасетах с шумами и без. После применяют обученную таким образом нейронную сеть для дальнейшей очистки реальных данных и повышения разрешающей способности сейсмической записи путём увеличения частотного диапазона реальных данных. Затем проводят анализ обработанных данных для ...

A system and method for selecting a service supplier

A computer system for selecting a service provider comprises a customer data processing module for receiving customer usage information and a machine learning module that comprises a trained data model which analyses the usage information and predicts future customer usage. The system further comprises a benchmarking module which compares the usage information with benchmarking data, a tender database which formats client tender information and creates an invitation to tender (ITT), a supplier database to which ITT responses are uploaded by suppliers, and a comparison module comparing and ranking the responses in accordance with one or more criteria and provides the ranking to the customer. The customer selects one of the responses supported by the usage information and ranking of responses. This allows for the customer to make better informed decisions about which supplier to use by using an automated system to match customer needs to the best supplier solution.

Classification of brain activity signals

A computer implemented method 100 of classifying brain activity signals 102 comprises receiving, as input to a neural network 104, input data comprising a plurality of brain activity signals 102A-D; applying a first block 108 to generate a plurality of first order wavelet scalograms 108A-D, then a first convolutional block 110A is configured to apply a plurality of Gabor filters, wherein each Gabor filter is associated with a learned bandwidth and learned frequency. Applying one or more further convolutional blocks 110B, 110C to the plurality of first order wavelet scalograms to generate a plurality of feature maps and applying a classification block 112 to the plurality of feature maps. The classification block is configured to generate one or more classifications of the plurality of brain activity signals from the plurality of feature maps. Applications include control of an artificial limb or a real or virtual vehicle.

Automatic assessment of histological indices

An apparatus for calculating a histological index from an image comprises a first neural network configured to receive the image as an input, to detect one or more inflammatory infiltrate regions in the image, and to output the detected inflammatory infiltrate regions as a first output. The apparatus further comprises a second neural network configured to receive the image as an input, detect one or more cell regions in the image and classify the detected cell regions wherein classifying the detected cell regions comprises determining whether each detected cell region corresponds to one or more predetermined cell types. The image is then filtered to remove detected cell regions which are determined not to correspond to one or more of the predetermined cell types and the filtered image is output as a second output. The apparatus also comprises a third neural network configured to receive the image as an input, to detect one or more granular tissue regions in the image, and to output the ...

Implementing a scatter function on a neural network accelerator

One or more vectors of indices are converted to sparse index tensors in a one-hot sparse format. An update tensor is generated, by applying the update values to one of the sparse index tensors (or a tensor derived from it). In some examples, an input data tensor is updated based on the update tensor. The update tensor may contain a plurality of input values to be updated. A sparse index tensor may be summed over one direction to calculate a reduced sparse index tensor, and element-wise multiplication performed. The neural network accelerator may use a delta function for clamp operation and a memory manipulation module to perform transpose, permute and repeat operations. This may be used in robotics or autonomous vehicles where Lidar produces a point cloud of object detection. A first and second vector of indices may relate to a first and second dimension, and combined tensors may produce a 2-D sparse index tensor. The scatter operation may be used in steganography or watermarking.

Scoring autonomous vehicle trajectories using reasonable crowd data

ARTIFICIAL INTELLIGENCE BASED DIGITAL LEASING ASSISTANT

A leasing AI platform receives a message pertaining to leasing a real estate unit, generates a token matrix representing at least a portion of the message, and provide the token matrix as an input to a trained machine learning model. An output of the trained machine learning model comprises an indication of a first category associated with the at least the portion of the message. The leasing AI platform identifies one or more actions associated with the first category, the actions pertaining to leasing the real estate unit, and automatically executes the one or more actions without human involvement in response to receiving the message.

SERVER AND CONTROL METHOD THEREOF

A server, a method for controlling thereof, and a method for controlling an electronic apparatus are provided. The method for controlling the server according to an embodiment includes: obtaining road information in a region having a predetermined range and information on a plurality of places in the region; obtaining a region mobility graph corresponding to the region, based on movement information of at least one user between the plurality of places, the region mobility graph including a plurality of nodes corresponding to the plurality of places and an edge connecting the plurality of nodes; learning the region mobility graph, by using a graph convolutional network (GCN) model for predicting a relationship between the plurality of nodes in the region mobility graph; and providing the learned region mobility graph to an external apparatus.

NEURAL NETWORK PROCESSING DEVICE, METHOD, AND COMPUTER-READABLE RECORDING MEDIUM

The 1 dimensional data generation means 92 generates 1 dimensional data, by setting the number of elements of each dimension other than predetermined one dimension to 1 based on multidimensional data corresponding to one input data. The number of elements reducing means 93 reduces the number of elements included in the 1 dimensional data. The copy means 95 generates multidimensional data, by copying the 1 dimensional data whose number of elements has been reduced multiple times. The convolution layer processing means 96 performs a convolution layer process with a filter size of 1×1 on the multidimensional data generated by the copy means 95. The element-wise product operation means 98 performs an element-wise product operation, based on the multidimensional data corresponding to one input data and multidimensional data generated by the above process.

SYSTEM AND METHOD FOR ESTIMATING HUMAN JOINT MOVEMENTS AND CONTROLLING EXOSKELETON ASSISTANCE

A device may include a high-level control layer comprising a convolutional neural network (CNN) configured to receive exoskeleton sensor data from one or more sensors on an exoskeleton and generate a user state estimate. The device may include a mid-level control layer configured to receive the user state estimate and generate a torque command for an actuator based on the user state estimate. The user state estimate may be an estimated gait phase, where the mid-level control layer generates the torque command as a function of the estimated gait phase based on an assistance profile. The high-level control layer may include a backward labeler and a real-time adaptation trainer. The backward labeler relabels ground truth gait phase from the exoskeleton sensor data using a local peak detection. The real-time adaptation trainer trains the CNN in a single epoch of backpropagation with the ground truth gait phase.

BATTERY CHARGING METHOD AND APPARATUS, AND DEVICE AND MEDIUM

Disclosed in the present application are a battery charging method and apparatus, and a device and a medium. The method includes; collecting, at a preset time interval, battery charging state data of a rechargeable battery within each of the preset time intervals in real time; when the battery charging state data within any one of the preset time intervals is collected, using the battery charging state data within the preset time interval as a training data set, inputting the training data set to an initial neural network model for training, and during the process of training, updating a preset network parameter on the basis of the difference value between a model output value corresponding to each piece of battery charging state data and a preset threshold until difference value between model output value corresponding to present battery charging state data and the preset threshold meets a preset error condition.

METHOD AND APPARATUS WITH NEURAL NETWORK ARCHITECTURE SEARCH

Disclosed is a method and apparatus for searching for an optimal architecture of a neural network. The apparatus may include a processor configured to generate a neural network loss based on parameters of a candidate architecture for the neural network, measure first hardware resources used in operation of the neural network with the candidate architecture, generate a prediction, using a hardware resource prediction model, of second hardware resources that would be used for operating the neural network with the candidate architecture, determine a hardware resource loss based on the first hardware resources and the second hardware resources, and determine a target architecture of the neural network based on the neural network loss and the hardware resource loss.

METHOD AND SYSTEM FOR MATERIAL DECOMPOSITION IN DUAL- OR MULTIPLE-ENERGY X-RAY BASED IMAGING

A method and system for generating material decomposition images from plural-energy x-ray based imaging, the method comprising: modelling spatial relationships and spectral relationships among the plurality of images by learning features from the plurality of images in combination and one or more of the plurality of images individually with a deep learning neural network; generating one or more basis material images employing the spatial relationships and the spectral relationships; and generating one or more material specific or material decomposition images from the basis material images. The neural network has an encoder-decoder structure and includes a plurality of encoder branches; each of one or more of the plurality of encoder branches encodes two or more images of the plurality of images in combination; and each of one or more of the plurality of encoder branches encodes a respective individual image of the plurality of images.

Система видеоинспекции плоских протяженных объектов

Изобретение относится к системам оптического обнаружения дефектов поверхности плоских протяженных объектов, таких как стальные металлические материалы (полосы, листы), их сплавы, композитные полимерные материалы и т.д. Система видеоинспекции плоских протяженных объектов содержит видеокамеру, систему освещения, систему управления оборудованием, ЭВМ со свёрточной нейронной сетью. Оптическая схема системы видеоинспекции содержит по меньшей мере одну RGB-камеру машинного зрения и комплексированную систему, состоящую из системы диффузионного освещения с по меньшей мере одним линейным осветителем и системы бокового освещения с по меньшей мере одним матричным осветителем. Технический результат - повышение качества детектирования дефектов поверхности плоских протяженных объектов. 1 ил.

METHOD AND APPARATUS FOR SUPPORTING DESIGN CO-WORK USING NEURAL NETWORKS

Embodiments provide a technique for assisting design collaboration using a neural network. According to embodiments, a method for assisting design collaboration comprises: an operation of transmitting login information of a user account to a server by a user terminal; an operation of authenticating the user account based on the login information by the server; an operation of generating a first design image according to a user input by the user terminal; an operation of receiving the first design image by the server; an operation of loading user information of the user account by the server; an operation of using a design generation model composed of a neural network based on the user information and the first design image and generating a second design image by the server; an operation of transmitting the second design image to a designer terminal by the server; an operation of transmitting feedback data for the second design image by a designer input to the server by the designer terminal ...

DYNAMIC PROFILE ASSIGNMENT AND ADJUSTMENT FOR CAMERA BASED ARTIFICIAL INTELLIGENCE OBJECT DETECTION

A system and method for improving the ability of a camera to detect objects or events occurring within its field of view and to accurately categorize them using Artificial Intelligence (AI) aided by input from users. The camera may include rules for determining when an object has entered its field of view, and for determining what category of object it is. When a new object is detected, an alert may be sent to a user and optionally to an analytics service as well. The user may provide input confirming whether the category of the event was correctly determined, and the analytics service may apply an AI algorithm to determine what, if any, changes should be made to the rule criteria in the camera. Updated rule criteria may be sent back to the camera thus improving its ability to detect objects in the future.

METHOD AND APPARATUS FOR PROGRAMMABLE AND CUSTOMIZED INTELLIGENCE FOR TRAFFIC STEERING IN 5G NETWORKS USING OPEN RAN ARCHITECTURES

A method of optimizing traffic steering (TS) radio resource management (RRM) decisions for handover of individual user equipment (UE) in Open Radio Access Network (O-RAN) includes: providing an O-RAN-compliant near real time RAN intelligent controller (near-RT RIC) configured to interact with O-RAN nodes; and utilizing an artificial intelligence (Al)-based TS application xApp in the near-RT RIC to optimize TS handover control and maximize UE throughput utility. The TS xApp is configured utilizing a virtualized and simulated environment for O-RAN, which virtualized and simulated environment for O-RAN is provided by ns-O-RAN platform. The optimization problem to be solved is formulated as a Markov Decision Process (MDP), and a solution to the optimization problem is derived by using at least one reinforcement learning (RL) technique.

AUTOMATED CREATION OF TINY DEEP LEARNING MODELS BASED ON MULTI-OBJECTIVE REWARD FUNCTION

State of art techniques existing method refer to handling multiple objectives such as accuracy and latency. However, the reward functions are static and not tunable at user end. Further, for NN search with hardware constraints, approaches combine various techniques such as Reinforcement learning, Evolutionary Algorithm (EA) etc., however hardly any work attempts to disclose combining different NAS approaches in unison to reduce the search space of other. Embodiments of the present disclosure provide a method and system for automated creation of tiny Deep Learning (DL) models to be deployed on a platform having a set of hardware constraints. The method performs a coarse-grained search using a Fast EA NAS model and then utilizes a fine-grained search to identify customized and optimized tiny model. The coarse-grained search and the fine-grained search performed by agents based on a weighted multi-objective reward function, which are tunable at user end.

NEURAL NETWORK COMPUTING METHOD AND APPARATUS, AND RELATED DEVICE

EFFICIENT DEEP LEARNING INFERENCE OF A NEURAL NETWORK FOR LINE CAMERA DATA

The invention discloses an automated method for accelerating deep learning inference of a neural network with layers, whereby a line-wise image consisting of pixels is generated by a line-camera (1) scanning an object (3), characterized by: for each new pixel-line added to the image, results of previously calculations for pixels of the current layer are used instead of repeated calculations to calculate the value of a pixel in the next layer. A corresponding arrangement comprising a neural network is disclosed as well.

Способ всеракурсного распознавания типового состава групповой воздушной цели из класса "самолеты с турбореактивными двигателями" на основе калмановской фильтрации и нейронных сетей

Изобретение относится к области вторичной цифровой обработки радиолокационных (РЛ) сигналов и может быть использовано для распознавания типового состава групповой воздушной цели (ГВЦ) из класса «самолеты с турбореактивными двигателями (ТРД)» при различных ракурсах ее РЛ наблюдения. Техническим результатом является возможность распознавания в импульсно-доплеровской радиолокационной станции с вероятностью, не ниже заданной, типового состава ГВЦ при различных ракурсах ее РЛ наблюдения. Заявленный способ основан на обработке амплитудно-частотного спектра РЛ сигнала, отраженного от ГВЦ из класса «самолеты с ТРД», спектральные составляющие которого обусловлены отражениями сигнала от планеров самолетов группы и вращающихся лопаток первых ступеней рабочих колес компрессоров низкого давления (КНД) их двигателей. Вычисляется процедура оптимальной калмановской фильтрации с соответствующими динамическими моделями по отсчетам доплеровских частот спектральных составляющих сигналов, отраженных от каждого ...

Operating a vehicle

Learnable Filters for EEG Classification

ENHANCED DYNAMIC RANDOM ACCESS MEMORY (EDRAM)-BASED COMPUTING-IN-MEMORY (CIM) CONVOLUTIONAL NEURAL NETWORK (CNN) ACCELERATOR

An enhanced dynamic random access memory (eDRAM)-based computing-in-memory (CIM) convolutional neural network (CNN) accelerator comprises four P2ARAM blocks, where each of the P2ARAM blocks includes a 5T1C ping-pong eDRAM bit cell array composed of 64×16 5T1C ping-pong eDRAM bit cells. In each of the P2ARAM blocks, 64×2 digital time converters convert a 4-bit activation value into different pulse widths from a row direction and input the pulse widths into the 5T1C ping-pong eDRAM bit cell array for calculation. A total of 16×2 convolution results are output in a column direction of the 5T1C ping-pong eDRAM bit cell array. The CNN accelerator uses the 5T1C ping-pong eDRAM bit cells to perform multi-bit storage and convolution in parallel. An S2M-ADC scheme is proposed to allot an area of an input sampling capacitor of an ABL to sign-numerical SAR ADC units of a C-DAC array without adding area overhead.

Detecting a defective nozzle in a digital printing system

A method includes, receiving a first digital image (FDI) to-be printed by a digital printing system (DPS) (10). In a in training phase: for first selected regions (111) in the FDI, a first set of synthetic images (SIs) (112A, 112B, 114A, 114B, 116A, 116B) having a defect caused by a defective part (DP) (99) in the first selected regions, is produced; a neural network (NN) (150) is trained to detect the defect using the first set SIs. In a subsequent detection phase: the NN is applied for identifying, in a second digital image (SDI) (136, 146) acquired from an image produced by the DPS, suspected second regions (135, 145); for each of the second regions, a second set (137, 147) of SIs having DPs that form the defects, is produced; and the DP is identified by comparing, in each of the second regions, between the SDI and the second set SIs.

MACHINE LEARNING MODELS FOR PATENT VALUATION

Methods, systems, and computer readable media for using machine learning models to determine predicted values of patent documents. In some examples, a method includes training, by at least one processor, a machine learning model to predict patent value based on unstructured text from training patents and, for each training patent, a measure of patent value. The method includes supplying, by the at least one processor, unstructured text from a patent document to the machine learning model. The method includes outputting, by the at least one processor, a predicted measure of value of the patent document.

INFORMATION PROCESSING DEVICE, INFORMATION PROCESSING SYSTEM, INFORMATION PROCESSING METHOD, AND RECORDING MEDIUM

An edge device (20) that is an example of an information processing device of an embodiment according to the present disclosure includes a transmitting section that transmits, to a server device (10) that generates a neural network, information related to a processing capability for processing the neural network supplied from the server device (10), wherein the information related to the processing capability includes at least one of capacity information of the neural network, filter size information of a favorite convolutional neural network, hardware architecture type information, chip information, and device model number information.

DEVICE AND METHOD FOR PROCESSING A CONVOLUTIONAL NEURAL NETWORK WITH BINARY WEIGHTS

Various embodiments relate to convolutional neural networks (CNN). CNN may be provided with a convolution kernel configured with binary weights. The CNN may be trained with the convolution kernel to determine a set of binary weights for the convolution kernel. The set of binary weights may be used for inference of the CNN. Devices, methods, and computer programs are disclosed.

SYSTOLIC NEURAL NETWORK ENGINE WITH CROSSOVER CONNECTION OPTIMIZATION

DEEP LEARNING METHOD FOR DEFECT CHARACTERIZATION

A method and system employing deep learning techniques improves processing speed and spatial resolution of acoustic wavenumber spectroscopy (ASSESS) techniques by performing semantic segmentation on simulated ultrasonic wavefield images of a steady-state, select-tone excitation of a structural or mechanical component. One or more embodiments may employ a convolutional neural network (CNN), pre-trained on openly-available datasets, and trained by transfer learning on an augmented wavefield dataset, to localize and characterize defects or damage from inspection measurements of components.

ASSISTIVE SMART GLASSES FOR VISUAL IMPAIRMENT, AND SYSTEM AND CONTROL METHOD THEREOF

The present invention discloses assistive smart glasses for visual impairment, and a system and a control method thereof. The smart glasses include a glasses body, wherein a microprocessor, a bone conduction headphone, a wireless communication module, a CCD optical module, a DSP image processing module, a camera lens, a memory module, a power supply module, a rechargeable battery, and an I/O data port are integrated in the glasses body; the camera lens is disposed at a front end of the glasses body; the CCD optical module converts an optical image captured by the camera lens into high-resolution compression data and sends the high-resolution compression data to the DSP image processing module; the microprocessor is electrically connected to the bone conduction headphone, the wireless communication module, and the DSP image processing module via the I/O data port; the memory module and the power supply module are electrically connected to the microprocessor; and the rechargeable battery ...

ACOUSTIC SYSTEM AND METHOD BASED GESTURE DETECTION USING SPIKING NEURAL NETWORKS

Conventional gesture detection approaches demand large memory and computation power to run efficiently, thus limiting their use in power and memory constrained edge devices. Present application/disclosure provides a Spiking Neural Network based system which is a robust low power edge compatible ultrasound-based gesture detection system. The system uses a plurality of speakers and microphones that mimics a Multi Input Multi Output (MIMO) setup thus providing requisite diversity to effectively address fading. The system also makes use of distinctive Channel Impulse Response (CIR) estimated by imposing sparsity prior for robust gesture detection. A multi-layer Convolutional Neural Network (CNN) has been trained on these distinctive CIR images and the trained CNN model is converted into an equivalent Spiking Neural Network (SNN) via an ANN (Artificial Neural Network)-to-SNN conversion mechanism. The SNN is further configured to detect/classify gestures performed by user(s).

SYSTEM AND METHOD FOR AUTOMATIC IDENTIFICATION OF LEGAL ENTITIES

Systems, methods, and computer readable media for identifying entities as legal entities are provided. These techniques may include accessing a corpus of documents and applying a persona prediction machine learning algorithm to classify entities associated with the corpus of documents. The persona prediction machine learning algorithm may include two layers. A first layer includes applying a signature block classifier that analyzes signature blocks of the entities. A second layer includes applying an entity classifier that analyzes a plurality of documents and/or network graphs associated with the entities. An entity database is updated to indicate the output of the persona prediction machine learning algorithm based on the signature block classifier and/or the entity classifier.

PAPER JAM INDICATION ESTIMATION DEVICE, PAPER JAM INDICATION ESTIMATION METHOD, AND RECORDING MEDIUM

A paper jam indication estimation device estimates an indication that a paper jam will occur in a paper feed device, and includes: a sound collector that collects a friction sound produced between sheets of paper when the paper is fed into the paper feed device from a holder holding a plurality of sheets of the paper; an estimator that, based on an output result obtained by inputting information pertaining to the friction sound into a trained model that is a machine learning model which has been trained, estimates a presence or absence of an indication that a paper jam will occur in the paper feed device; and an outputter that, when the estimator estimates that the indication that a paper jam will occur is present, outputs, to the paper feed device, a signal that stops the paper from being fed into the paper feed device.

METHOD FOR GENERATING MATRIX DATA FOR CONVOLUTIONAL NEURAL NETWORK AND LEARNING SYSTEM USING CONVOLUTIONAL NEURAL NETWORK

In a method for generating matrix data for a convolutional neural network that performs a convolution operation based on the matrix data in which vehicle information is arranged as an matrix element, the matrix data is composed of predetermined time-series data in which each row changes continuously in terms of time in an arrangement direction of each column, the time-series data is composed of first data of which degree of influence on the convolution operation is high and second data of which degree of influence is lower than the first data, the convolution operation is performed using a kernel that partitions the matrix data into the rows and columns corresponding to a predetermined coefficient, and at least one row of the first data is arranged for each set of rows corresponding to the coefficient.

COMPUTER-IMPLEMENTED METHOD, COMPUTER PROGRAM PRODUCT AND SYSTEM FOR DATA ANALYSIS

Способ детектирования и дисперсионного анализа частиц в текучей среде с применением метода искусственного интеллекта

Изобретение относится к области детектирования частиц в текучей среде с последующим анализом их размера и может быть использовано в тех областях науки и техники, где требуется анализ микроструктуры эмульсий, суспензий и золей, например, в нефтехимической, фармацевтической и пищевой промышленностях, медицинских исследованиях и биохимии, метеорологии, а также в системах контроля и защиты окружающей среды. Технический результат заключается в повышении точности измерений, а также в повышении быстродействия при снижении объема требуемых вычислительных ресурсов. Способ дисперсионного анализа частиц в текучей среде предусматривает использование нейронной сети с бинарными сверточными слоями, в которой для каждого сверточного слоя матрицы входных признаков, соответствующих своему входному каналу, фильтруются посредством покомпонетного умножения на каждый из базовых шаблонов с получением матриц, к которым покомпонетно применяют сигмоидальную нелинейную функцию, после чего покомпонентно складывают ...

Structural variant identification

Methods are disclosed for analyzing sequence data to identify putative structural variants (SVs), filter out germline SVs and artifacts from sample handling and sequencing to leave only somatic SVs. Recognizing that those somatic SVs, especially where sequenced from a tumor or other sample of diseased tissue, may be indicative of the presence of disease, methods may include designing primers to selectively amplify those somatic SVs for monitoring disease progression or recurrence in patient samples including blood. In various embodiments, the original sequence data may be obtained from FFPE-extracted or fresh frozen-extracted DNA and somatic SVs may be identified without the benefit of a matched normal sequence. In some embodiments, machine learning analysis may be used in the identification of SVs, the filtering of artifacts and germline SVs, and/or primer and probe design for disease monitoring.

SYNTHETIC CT USING MRI

Disclosed are various approaches for generating synthetic computer tomography (CT) images using magnetic resonance imaging A patient body outline (PBO) of a patient is obtained. Then, a magnetic resonance image set in a limited field of view (MRI-in-LFOV) of the patient is converted into a synthetic computed tomography (CT) image set in limited field of view (syn-CT-in-LFOV) of the patient. Next, the syn-CT-in-LFOV is expanded to a synthetic CT image set in full field of view (syn-CT-in-FFOV) of the patient based at least in part on the PBO.

Battery charging method and apparatus, and device and medium

Disclosed in the present application are a battery charging method and apparatus, and a device and a medium. The method includes; collecting, at a preset time interval, battery charging state data of a rechargeable battery within each of the preset time intervals in real time; when the battery charging state data within any one of the preset time intervals is collected, using the battery charging state data within the preset time interval as a training data set, inputting the training data set to an initial neural network model for training, and during the process of training, updating a preset network parameter on the basis of the difference value between a model output value corresponding to each piece of battery charging state data and a preset threshold until difference value between model output value corresponding to present battery charging state data and the preset threshold meets a preset error condition.

Anomaly Detection System for Embedded Devices

An anomaly detection system may be configured for an embedded device, such as a microcontroller. The anomaly detection system may be configured to receive a dataset from a sensor, such as a camera, a microphone, or an inertial management unit. The anomaly detection system may extract a plurality of features from the dataset. The plurality of features may be configured to train a neural network model, such as a convolutional classifier, to generate one or more classifications. The anomaly detection system may generate an anomaly score based on the plurality of features. The anomaly detection system may trigger an output based on the anomaly score exceeding a range.

RADIO FREQUENCY SYSTEM INCLUDING RECOMMENDATION TRAINING AGENT FOR MACHINE LEARNING ALGORITHM AND RELATED METHODS

A radio frequency (RF) system may include at least one RF sensor in an RF environment and at least one RF actuator. The RF system may also include at least one processor that includes a machine learning agent configured to use a machine learning algorithm to generate an RF model to operate the at least one RF actuator based upon the at least one RF sensor. The processor may also include a recommendation training agent configured to generate performance data from the machine learning agent, and change the RF environment based upon the performance data so that the machine learning agent updates the machine learning algorithm.

END-TO-END WATERMARKING SYSTEM

STACKED CONVOLUTIONAL LONG SHORT-TERM MEMORY FOR MODEL-FREE REINFORCEMENT LEARNING

WORKFLOW PREDICTIVE ANALYTICS ENGINE

Systems, methods, and apparatus to generate and utilize predictive workflow analytics and inferencing are disclosed and described. An example apparatus includes memory circuitry including instructions and a plurality of artificial intelligence (AI) models; and processor circuitry to execute the instructions to implement at least: a smart scheduling engine to train at least one of the plurality of AI models, update at least one of the plurality of AI models, and inference a prediction using at least one of the plurality of AI models; and a smart scheduling application programming interface (API) to facilitate interaction with at least one of the plurality of AI models to trigger the prediction and to configure resources for an appointment based on the prediction.

Transfer learning-based method for improved VGG16 network pig identity recognition

Disclosed is a transfer learning-based method for improved VGG16 network pig identity recognition. The method comprises: first performing frame by frame extraction on a processed video to obtain a series of pictures, which are preprocessed into a data set, and then dividing a training set and a test set; constructing an improved VGG16 network training model BN-VGG16; and saving a pre-trained feature extraction model Pre-VGG16; next is a transfer learning process: transferring a Pre-VGG16 feature extraction network obtained by source domain training to a Pig-VGG16 network for recognizing pigs; and performing multi-block improve absolute difference local direction pattern (MB-IADLDP) feature extraction on a data set that has undergone size adjustment, and performing serial fusion, and finally performing identity recognition on a pig. A transfer learning-based improved VGG16 model is superior to conventional VGG16 network models in terms of operating speed and precision.

Neural network architecture search apparatus and method

The disclosure relates to neural network architecture searching. Categorised training images are paired within their categories. A group convolution is then performed on each image pair P to generate a feature map FM for each image pair. The feature maps are input to a supernet to perform the neural network architecture search. Images may be paired by concatenating two images within the same category, with each image being concatenated only once. Each image has size H1×W1×IC1 where IC1 is the number of channels in the image, e.g. three for red, green, and blue. The group convolution may comprise convolving the first image of each pair with a first filter F1 of size H2×W2×IC1. Similarly, the second image of each pair may be convolved with a second filter F2 of the same size. Each of the first and second filters may output half the features in each feature map. The original, unpaired images may undergo a normal convolution to generate second feature maps used by a second supernet to train ...

PROCESSING METHOD IN A CONVOLUTIONAL NEURAL NETWORK ACCELERATOR, AND ASSOCIATED ACCELERATOR

A processing method in a convolutional neural network accelerator includes an array of unitary processing blocks associated with a set of respective local memories and performing computing operations on data stored in its local memories, wherein: during respective processing cycles, some unitary blocks receive and/or transmit data from or to neighbouring unitary blocks in at least one direction selected, on the basis of the data, from among the vertical and horizontal directions in the array; during the same cycles, some unitary blocks perform a computing operation in relation to data stored in their local memories during at least one previous processing cycle.

DISEASE PREDICTION METHOD AND APPARATUS

Provided is a disease prediction method and apparatus. The disease prediction apparatus may receive a medical image and clinical information, identify, from the medical image, quantitative data comprising a volume of at least one anatomical structure, and predict a disease occurrence based on the clinical information and the quantitative data. An artificial intelligence model may be used for each of identification of an anatomical structure and prediction of the disease occurrence. The disclosure was supported by the “AI Precision Medical Solution (Doctor Answer 2.0) Development” project hosted by Seoul National University Bundang Hospital (Project Serial No.: 1711151151, Project No.: S0252-21-1001).

ACCELERATION OF 2D DILATED CONVOLUTION FOR EFFICIENT ANALYTICS

Disclosed herein are improved systems and methods for accelerated 2D dilated convolution. A processor determines an offset based on a dilation factor of the 2D dilated convolution. The processor selects rows of data from the 2D input in phases based on the offset and loads an input feature panel without overwriting data that has not yet been consumed by the 2D dilated convolution processor. As the 2D dilated convolution processor performs the convolution iterations, the processor continues to load additional data for the convolution. As the convolution iterations are completed, the processor spaces result of the 2D dilated convolution into a matrix such that results of each phase are spaced based on the offset.

METHOD, DEVICE, AND COMPUTER PROGRAM PRODUCT FOR PROCESSING VIDEO

Methods, devices and computer program products for processing video are disclosed herein. A method includes: generating, based on a reference image and a first frame of a video comprising an object, a two-dimensional avatar image of the object; and generating a base three-dimensional avatar of the object by performing a three-dimensional transformation on the two-dimensional avatar image and the object in the first frame. The method further includes: generating a three-dimensional avatar video corresponding to the video based on the base three-dimensional avatar and features of the video, the features comprising differences of the object between adjacent frames of the video. This solution enables the generation of a customized three-dimensional avatar video 10 for an object in a video, where the avatar can move in synchronization with the object and retain the unique features of the object, and can provide a more detailed and vivid representation than a two-dimensional avatar.

PERFORMING AVERAGE POOLING IN HARDWARE

Respiration analysis method based on electrical activities of the diaphragm

... 본 발명은 인공신경망을 이용한 횡격막 전기적 활동 기반 호흡 분석 방법에 관한 기술로, 인공호흡기를 연결한 미숙아의 호흡 모니터링 중 실제 신경 호흡을 분석하기 위한 딥러닝을 이용한 호흡 검출 방법에 관한 기술이다. 구체적으로 미숙아의 횡격막 전기적 활동 신호(Edi, Electrical activity of diaphragm)에 포함된 불규칙한 노이즈를 제거하고 들숨과 날숨으로 구성된 단일 호흡 내에서 정점을 찾기 위하여, 국부 최대값(Local Maximum, LM)을 사용하여 후보 정점을 최대한 많이 검출하고, CNN 기반 심층 신경망(Deep Neural Network, DNN)을 사용하여 과잉 검출된 Edi 정점을 제거함으로써 최종 Edi 정점을 선택하는 2단계 접근 방식을 포함하는 방법에 관한 것이다.

SYSTEM, METHOD, AND COMPUTER READABLE STORAGE MEDIUM FOR AUTO-REGRESSIVE WAVENET VARIATIONAL AUTOENCODERS FOR ALIGNMENT-FREE GENERATIVE PROTEIN DESIGN AND FITNESS PREDICTION

A system, computer readable storage medium and method for generating protein sequences, includes an encoder configured to encode a plurality of input protein sequences onto a latent space distribution, and an autoregressive generator configured to decode the latent space distribution to generate new protein sequences different from the input protein sequences. The system is trained with a loss function that includes reconstruction loss and aa mutual information maximization term.

METHOD OF UPDATING A VELOCITY MODEL OF SEISMIC WAVES IN AN EARTH FORMATION

A method involving automated salt body boundary interpretation employs multiple sequential supervised machine learning models which have been trained using training data. The training data may consist of pairs of seismic data and labels as determined by human interpretation. The machine learning models are deep learning models, and each of the deep learning models is aimed to address a specific challenge in the salt body boundary detection. The proposed approach consists of application of an ensemble of deep learning models applied sequentially, wherein each model is trained to address a specific challenge. In one example an initial salt boundary inference as generated by a first trained first deep learning model is subject to a trained refinement deep learning model for false positives removal.

COMPUTER-IMPLEMENTED METHOD FOR PROVIDING AN ENCRYPTED DATASET PROVIDING A GLOBAL TRAINED FUNCTION, COMPUTER-IMPLEMENTED METHOD FOR RECOVERING PERSONAL INFORMATION, COMPUTER SYSTEM AND COMPUTER PROGRAM

Respective local parameters parametrizing a base function are determined to provide at least one local trained function for each of multiple client systems by training the respective local trained function using machine learning with multiple training datasets on the respective client system, wherein at least some of the training datasets are specific to the respective client system. A respective local plaintext dataset including the local parameters of the respective local trained function is encrypted to generate a respective local encrypted dataset on the respective client system. The local encrypted datasets are transmitted to an aggregating system, and the global encrypted dataset is calculated from the local encrypted datasets using the calculation algorithm.

TRAINING FOR IMAGE SIGNAL PROCESSING ALGORITHM ITERATIONS FOR AUTONOMOUS VEHICLES

In one embodiment, a system generates an image using either a first or a second image signal processing (ISP) algorithm, where the first or second ISP algorithm is applied to raw image data of a camera of an autonomous driving vehicle (ADV) to generate the image. The system applies a machine learning model to the image to identify a representation of an obstacle, where the machine learning model is generated by a few shots learning algorithm that contrasts labeled data of a positive training sample from images corresponding to the first and second ISP algorithms to labeled data of a negative training sample from images corresponding to the first and second ISP algorithm. The system determines a classification and a location of the obstacle based on the representation of the obstacle. The system plans a motion control of the ADV based on the classification and location of the detected object.

METHOD FOR TRANSFORMING A TRAINED NEURAL NETWORK

Selon un aspect, il est proposé un procédé de transformation d'un réseau de neurones artificiels entraîné (TNN) comportant une couche de convolution binaire (CNV_b) suivie d'une couche de mise en commun (PL) puis d'une couche de normalisation par lot (BNL), le procédé comprenant : - une obtention (10) du réseau de neurones artificiels entraîné (TNN), puis - une conversion (11) du réseau de neurones artificiels entraîné (TNN) dans laquelle l'ordre des couches du réseau de neurones artificiels entraîné (TNN) est modifié en déplaçant la couche de normalisation par lot après la couche de convolution (CNV), de façon à obtenir un réseau de neurones artificiels transformé (ONN).

Three-dimensional intelligent filling construction management method of subgrade assisted by mixed reality technology

The application provides a three-dimensional intelligent filling construction management method of subgrade assisted by mixed reality technology, which comprises the following steps: obtaining an MR three-dimensional realistic model; associating the absolute three-dimensional coordinates of the filling surface with the MR three-dimensional realistic model, and filling the subgrade based on an association result; obtaining a compaction degree of the filling subgrade based on the scanning characteristic image; obtaining a simplified calculation model based on the subgrade solid model, and inputting variable parameters of the filling subgrade into the simplified calculation model to obtain deformation data of the filling subgrade; and judging whether the subgrade deformation control index meets the design requirements based on the porosity, compaction degree and deformation data of filling subgrade.

A method for predicting at least one motion of at least one object in the surroundings of a motor vehicle as well as a corresponding detection system

The invention relates to predicting the motion of an object surrounding a vehicle, for possible use in autonomous vehicles. The method comprising: capturing the surroundings of the motor vehicle (12) by at least one camera and by at least one radar sensor and/or one lidar sensor; generating one grid map image (20); generating at least one RGB image (22), at least one semantic segmentation image (24), and at least one depth image (26); feeding the at least one grid map image (20) to a first convolutional long short-term memory network (14) for predicting one further grid map image (28); feeding the RGB image (22), the semantic segmentation image (24), and the depth image (26) to a second convolutional long short-term memory network (16) and appending the further grid map image (28) to hidden layers (Fig. 2; 30) of the second convolutional long short-term memory network (16) for predicting one further RGB image (32), one further semantic segmentation image (34), and one further depth image ...

Gearbox fault diagnosis model training method and gearbox fault diagnosis method

Disclosed in the present invention are a gearbox fault diagnosis model training method and a gearbox fault diagnosis method. The training method comprises: obtaining a motor current signal in an electromechanical system where a gearbox is located; calculating feature values representing the complexity and the mutation degree of the current signal according to the current signal; screening the feature values according to a random forest algorithm to generate a sample data set; and training a deep reinforcement learning network model according to the data set to generate a fault diagnosis model. According to the gearbox fault diagnosis model training method provided by the present invention, only the current signal is obtained, no additional sensor is needed, and the defect in the prior art that hardware is added is overcome. Feature data related to the fault is extracted by calculating and screening the feature values representing the complexity and the mutation degree of the current signal ...

반도체 장치 및 전자 기기

... 낮은 소비 전력으로 합성곱 처리를 수행할 수 있는 반도체 장치를 제공한다. 제 1 회로가 제 1 유지부와 제 1 트랜지스터를 가지고, 제 2 회로가 제 2 유지부와 제 2 트랜지스터를 가지는 반도체 장치이다. 제 1 회로 및 제 2 회로는 제 1 입력 배선과, 제 2 입력 배선과, 제 1 배선과, 제 2 배선에 전기적으로 접속되어 있다. 제 1 유지부는 제 1 트랜지스터에 흐르는 제 1 전류를 유지하는 기능을 가지고, 제 2 유지부는 제 2 트랜지스터에 흐르는 제 2 전류를 유지하는 기능을 가진다. 또한, 제 1 전류 및 제 2 전류는 합성곱 처리에 사용되는 필터값에 따라 결정된다. 제 1 입력 배선 및 제 2 입력 배선에 합성곱 처리가 수행되는 화상 데이터에 따른 전위가 입력됨으로써, 제 1 회로는 제 1 배선 및 제 2 배선 중 한쪽에 전류를 출력하고, 제 2 회로는 제 1 배선 및 제 2 배선 중 다른 쪽에 전류를 출력한다. 제 1 회로 및 제 2 회로가 제 1 배선 또는 제 2 배선에 출력하는 전류의 양은 필터값과 화상 데이터에 따라 결정된다.

뉴럴 네트워크 처리 장치, 정보 처리 장치, 정보 처리 시스템, 전자 기기, 뉴럴 네트워크 처리 방법 및 프로그램

... 메모리량을 삭감한다. 뉴럴 네트워크 처리 장치는, 계수 행렬에서의 값이 제로인 제1 계수의 위치를 제1 값으로 나타내고, 상기 계수 행렬에서의 값이 제로가 아닌 제2 계수의 위치를 제2 값으로 나타내는 제로 계수 위치 테이블과, 상기 계수 행렬에서의 상기 제2 계수를 유지하는 비제로 계수 테이블에 부호화된 상기 계수 행렬을 복호하는 복호부(41)와, 상기 복호부에 의해 복호된 상기 계수 행렬과, 변수 행렬의 컨벌루션 처리를 실행하는 곱합 회로(116)를 구비하고, 상기 복호부는, 상기 제로 계수 위치 테이블에서의 상기 제2 값으로 나타난 위치에 상기 비제로 계수 테이블에 격납된 상기 제2 계수를 격납함으로써, 상기 계수 행렬을 복호한다.

Video retrieval method and apparatus, device and storage medium

A video retrieval method and apparatus, a device and a storage medium are provided. The method comprises the following steps: acquiring a comparison video clip from a video library according to the duration of a to-be-tested video (S110); determining the similarity between the to-be-tested video and the comparison video clip by a target spatio-temporal neural network, a spatio-temporal convolutional layer of the target spatio-temporal neural network being configured to be capable of performing two-dimensional convolution and temporal dimension information processing, respectively (S120); and traversing the video library, and outputting a retrieval result according to the similarity (S130).

Method for predicting delay at multiple corners for digital integrated circuit

Disclosed in the present invention is a method for predicting a delay at multiple corners for a digital integrated circuit, which is applicable to the problem of timing signoff at multiple corners. In the aspect of feature engineering, a path delay relationship at adjacent corners is extracted by using a dilated convolutional neural network (Dilated CNN), and learning is performed by using a bi-directional long short-term memory model (Bi-directional Long Short-Term Memory, BLSTM) to obtain topology information of a path. Finally, prediction results of a path delay at a plurality of corners are obtained by using an output of a multi-gate mixture-of-experts network model (Multi-gate Mixture-of-Experts, MMoE). Compared with a conventional machine learning method, the present invention can achieve prediction with higher precision through more effective feature engineering processing in a case of low simulation overheads, and is of great significance for timing signoff at multiple corners of ...

Systems and Methods for Authenticating Jewelry and/or Gemstones

A variety of techniques for authenticating jewelry and gemstones is provided. With some examples, a stored reference signature for an item is accessed, where the item comprises jewelry and/or a gemstone, where the reference signature comprises feature values for a set of features about the item, and where the feature values are derived from previously sensed physical characteristic data about the item. A test item is sensed to generate sensed physical characteristic data about the test item. This sensed physical characteristic data about the test item is processed to extract new feature values for the set of features, where the new feature values serve as a test signature of the test item. One or more processors can compare the test signature with the accessed reference signature and authenticate the test item as being the same as the item if the comparing indicates that the test signature and reference signature match.

METHOD OF DETERMINING ZONE MEMBERSHIP IN ZONE-BASED FEDERATED LEARNING

A processor-implemented method includes receiving, by a user equipment (UE), a zone determination function based on registering for a federated learning process for training a first federated learning model. The method also includes determining, by the UE, a zone membership in accordance with UE parameters and the zone determination function. The method further includes selecting the first federated learning model, by the UE, based on the zone membership. The method includes training the first federated learning model by the UE.

THREE-DIMENSIONAL FLUID REVERSE MODELING METHOD BASED ON PHYSICAL PERCEPTION

A three-dimensional fluid reverse modeling method based on physical perception. The method comprises: encoding a fluid surface height field sequence by a surface velocity field convolutional neural network to obtain a surface velocity field at a time t; inputting the surface velocity field into a pre-trained three-dimensional convolutional neural network to obtain a three-dimensional flow field, wherein the three-dimensional flow field includes a velocity field and a pressure field; inputting the surface velocity field into a pre-trained regression network to obtain fluid parameters; and inputting the three-dimensional flow field and the fluid parameters into a physics-based fluid simulator to obtain a time series of the three-dimensional flow field. The requirements for real fluid reproduction and physics-based fluid reediting are met.

NON-CONTACT FATIGUE DETECTION SYSTEM AND METHOD BASED ON RPPG

The present disclosure provides a non-contact fatigue detection system and method based on rPPG. The system and method adopt multi-thread synchronous communication for real-time acquisition and processing of rPPG signal, enabling fatigue status detection. In this setup, the first thread handles real-time rPPG data capture, storage and concatenation, while the second thread conducts real-time analysis and fatigue detection of rPPG data. Through a combination of skin detection and LUV color space conversion, rPPG raw signal extraction is achieved, effectively eliminating interference from internal and external environmental facial noise; Subsequently, an adaptive multi-stage filtering process enhances the signal-to-noise ratio, and a multi-dimensional fusion CNN model ensures accurate detection of respiration and heart rate. The final step involves multi-channel data fusion of respiration and heartbeats, succeeding in not only learning person-independent features for fatigue detection but ...

VECTOR COMPUTATION UNIT IN A NEURAL NETWORK PROCESSOR

A circuit for performing neural network computations for a neural network comprising a plurality of layers, the circuit comprising: activation circuitry configured to receive a vector of accumulated values and configured to apply a function to each accumulated value to generate a vector of activation values; and normalization circuitry coupled to the activation circuitry and configured to generate a respective normalized value from each activation value.

CNN PROCESSING METHOD AND DEVICE

RADIOLOGICALLY NAVIGATED INTRA-OPERATIVE SPINAL GUIDANCE SYSTEM

Data Processing Method, System and Device, and Readable Storage Medium

A data processing method, system and device, and a readable storage medium. The method includes: marking each layer of a network model as a key layer or a non-key layer according to acquired structural information of the network model; respectively determining a quantization bit width range of the key layer and a quantization bit width range of the non-key layer according to hardware resource information that needs to be deployed; determining, in the quantization bit width range, optimal quantization bit widths of each layer of the network model; and training the network model based on the optimal quantization bit widths of each layer of the network model, so as to obtain an optimal network model, and performing data processing using the optimal network model. According to the present disclosure, for an optimal network model obtained by means of performing training based on an optimal quantization bit width, insofar as the optimal accuracy of the network model is ensured, the model structure ...

ELECTRONIC APPARATUS AND METHOD FOR CONTROLLING THEREOF

An electronic apparatus and a method for controlling thereof are provided. The electronic apparatus includes a memory storing an artificial neural network and metadata including information of at least one layer in the artificial neural network, and a processor configured to: acquire a security vector based on the metadata and a security key of the electronic apparatus; map the security vector and the metadata with the security key and identification information of the artificial neural network; perform encryption on the at least one layer based on the metadata and the security vector; based on input data input to the artificial neural network, load the metadata and the security vector by using the security key and the identification information of the artificial neural network; and perform an operation between the input data and the encrypted at least one layer based on the loaded security vector and the metadata.

MACHINE LEARNING SYSTEM, RECOGNIZER, LEARNING METHOD, AND PROGRAM

There are provided a machine learning system, a learning method, and a program that can facilitate learning of a large number of combined medical images. A machine learning system (10) includes: an image database (14) that stores a plurality of medical images; a geometric-shape database (16) that stores geometric shapes to be superimposed on the medical images; a processor (22); and a learning model (30), in which the processor (22) is configured to perform a selection accepting process of accepting selection of a medical image from among the medical images stored in the image database (14) and selection of a geometric shape from among the geometric shapes stored in the geometric-shape database (16), a geometric-shape combining process of combining the selected medical image and the selected geometric shape and generating a composite image, and a training process of making the learning model perform learning by using the composite image.

Artificial intelligence using convolutional neural network with hough transform

Artificial intelligence using convolutional neural network with Hough Transform. In an embodiment, a convolutional neural network (CNN) comprises convolution layers, a Hough Transform (HT) layer, and a Transposed Hough Transform (THT) layer, arranged such that at least one convolution layer precedes the HT layer, at least one convolution layer is between the HT and THT layers, and at least one convolution layer follows the THT layer. The HT layer converts its input from a first space into a second space, and the THT layer converts its input from the second space into the first space. The CNN may be applied to an input image to perform semantic image segmentation, so as to produce an output image representing a result of the semantic image segmentation.

RECONFIGURABLE ARCHITECTURE FOR FUSED DEPTH-WISE SEPARABLE CONVOLUTION (DSC)

A method of operating a depth-wise separable convolutional (DSC) network on a DSC accelerator includes determining a difference between a first throughput associated with a depth-wise convolution (DWC) engine of the DSC accelerator and a second throughput associated with a point-wise convolution (PWC) engine of the DSC accelerator. The method also includes selectively activating, for each layer of the DSC network, each first processing elements (PEs) in one or more of a first set of columns of first PEs associated with the DWC engine and/or each second PE in one or more of a second set of columns associated with the PWC engine based on the difference between the first throughput and the second throughput. The method further includes processing, for each layer of the DSC network, an input via the DSC accelerator based on selectively activating each first PE and/or each second PE.

METHODS AND SYSTEMS FOR EXECUTING A NEURAL NETWORK ON A NEURAL NETWORK ACCELERATOR

Methods of dividing a neural network comprising one or more layers into chunks of operations executable in a hardware pass of hardware configurable to execute a neural network. The one or more layers of the neural network are divisible into one or more layer groups that comprise a sequence of layers executable in the same hardware pass of the hardware. Each layer group is divisible into one or more chunks of operations executable in a hardware pass of the hardware. The one or more chunks for a layer group are defined by one or more split parameters. The method includes: obtaining a layer group loss function that represents a performance metric associated with executing a layer group on the hardware as a function of the one or more split parameters and one or more neural network architecture parameters for the layer group; generating a neural network loss function based on the layer group loss function that represents the performance metric associated with executing the neural network on ...

SYSTEMS AND METHODS FOR PROBABILISTIC FORECASTING OF EXTREMES

A computer-implemented method for producing probabilistic forecasts of extreme values. The method comprises obtaining input data comprising a plurality of signals of interest and a plurality of covariates associated therewith, each covariate of the plurality of covariates having an associated data type. The method further comprises performing a first forecast based on the input data. Performing the first forecast comprises: obtaining one or more trained machine learning models, each trained machine learning model of the one or more trained machine learning models having been trained to map one or more covariates of a respective data type to one or more surrogate covariates; mapping, using the one or more trained machine learning models and the input data, the plurality of covariates to one or more surrogate covariates, the one or more surrogate covariates corresponding to a compressed representation of the input data; fitting a statistical model of extremes to the plurality of signals of ...

METHOD AND DEVICE FOR THE TRAINING OF A MACHINING LEARNING ROUTINE FOR CONTROLLING A TECHNICAL SYSTEM

IMAGE COMPONENT GENERATION BASED ON APPLICATION OF ITERATIVE LEARNING ON AUTOENCODER MODEL AND TRANSFORMER MODEL

An electronic device and method for image component generation based on application of iterative learning on autoencoder model and transformer model is provided. The electronic device fine-tunes, based on first training data including a first set of images, an autoencoder model and a transformer model. The autoencoder model includes an encoder model, a learned codebook, a generator model, and a discriminator model. The electronic device selects a subset of images from the first training data. The electronic device applies the encoder model on the selected subset of images. The electronic device generates second training data including a second set of images, based on the application of the encoder model. The generated second training data corresponds to a quantized latent representation of the selected subset of images. The electronic device pre-trains the autoencoder model to create a next generation of the autoencoder model, based on the generated second training data.

System and method for diagnosing small bowel cleanliness

The present invention relates to a system for diagnosing small bowel cleanliness. The system may comprise: a similarity analysis unit for analyzing to select a representative image of similar small bowel images from among a plurality of small bowel images; an image classification unit for, when a series of a plurality of small bowel images in which cleanliness is to be diagnosed are received in a state where the plurality of small bowel images have been learned, classifying small bowel cleanliness according to scores by predicting the small bowel cleanliness by applying the representative image to a result of the learning; and a cleanliness diagnosis unit for calculating final small bowel cleanliness for the series of the plurality of small bowel images on the basis of a score for small bowel cleanliness of the representative image and the number of small bowel images similar to the representative image.

Implementation of discrete Fourier-related transforms in hardware

Discrete Fourier transforms are carried out by fixed function circuitry to perform convolution operations. Convolution kernel derived from a weight matrix representing a matrix multiplication may be used to execute a discrete Fourier related transform. Input data containing a real value input tensor may undergo a fast Fourier transform. Another dimension of the input tensor may be Fourier transformed to produce a fast Fourier transform of the input data. A short time Fourier transform may be implemented using convolution kernels where strided convolution operations correspond to an amount of overlap between overlapping parts of input data. The convolution kernel may be generated by reshaping of permutating dimensions of a weight matrix, before input data is obtained. A second tensor comprising imaginary parts of values may also undergo discrete Fourier transform, and a concatenated tensor produced. Further splitting the convolution output may provide further multiplied tensors.

Weakly-supervised sound event detection method and system based on adaptive hierarchical pooling

The present disclosure provides a weakly-supervised sound event detection method and system based on adaptive hierarchical pooling. The system includes an acoustic model and an adaptive hierarchical pooling algorithm module, where the acoustic model inputs a pre-processed and feature-extracted audio signal, and predicts a frame-level prediction probability, and the adaptive hierarchical pooling algorithm module aggregates the frame-level prediction probability to obtain a sentence-level prediction probability. The acoustic model and a relaxation parameter are jointly optimized to obtain an optimal model weight and an optimal relaxation parameter, and an optimal pooling strategy is formulated for each category of sound event based on the optimal relaxation parameter. A pre-processed and feature-extracted unknown audio signal is input to obtain frame-level prediction probabilities of all target sound events, and sentence-level prediction probabilities of all categories of target sound events ...

Pose estimation-based pedestrian fall action recognition method and device

The present application provides a pose estimation-based pedestrian fall action recognition method and device. According to the present application, a multi-scale adjacency matrix is used to realize information aggregation, and residual connection is introduced between upper and lower spatial-temporal combining modules having a same structure; spatial-temporal combined features of a pose in double flows (a key point flow and a bone edge flow) are respectively extracted; and finally the double-flow results are combined to make a fall action determination, so that the influence of a background on a recognition effect is reduced to improve action recognition accuracy, and the computational complexity is reduced.

Method for recognizing type of vortex signal of evaporator of nuclear power plant on basis of LSTM-CNN

A method for recognizing the type of a vortex signal of an evaporator of a nuclear power plant on the basis of an LSTM-CNN. The method comprises: calibrating data of each channel of a vortex signal; processing the calibrated data by using a time window; processing time sequence data in a differential manner; extracting time feature information of a time sequence by means of an LSTM network; a CNN network extracting local feature information of the time sequence; fusing the feature information of the LSTM network and that of the CNN network, wherein after the training and learning of a large amount of data, the feature information thereof can be represented by means of an input signal in vector form by using a triple loss principle; and constructing a defect signal feature database, representing same in vector form, comparing same to obtain the Euclidean distance between the vector feature of the input signal and the vector feature in a defect library, determining, according to the magnitude ...

SYSTEM AND METHOD FOR MANUFACTURING QUALITY CONTROL USING AUTOMATED VISUAL INSPECTION

A system and method for automated visual inspection is provided herein. The method includes providing an inspection image of the article to an object detection model trained to detect at least one defect type in an input image and generating object location data identifying a location of a detected object in the inspection image; comparing the inspection image to a golden sample image to identify an artifact in the inspection image corresponding to a difference between the inspection image and the golden sample image, wherein the artifact is defined by artifact location data describing a location of the artifact in the inspection image; and determining whether the artifact location data matches the object location data according to predetermined match criteria.

TENSOR TRANSFER THROUGH INTERLEAVED DATA TRANSACTIONS

A device includes a direct memory access (DMA) controller comprising DMA channels, a bridge circuit configured to couple the DMA channels to memory channels coupled to respective memory modules, and a local memory unit. The DMA controller is configured to transfer tensor data between the local memory unit and the memory modules via the DMA channels and the memory channels using concurrent data transactions, the tensor data is stored and addressed as parts of a single tensor in the local memory unit, and the tensor data is interleaved onto the memory modules and is stored and addressed as sub-tensors in respective memory modules.

Performing a processing task instructed by an application

A computer-implemented method of performing a processing task instructed by an application executing on a computing device, wherein the processing task generates a task output based on a task input. The method obtains (302) a neural network trained to perform the processing task and comprising a plurality of intermediate output layers. The method further obtains (304) a decision-making data structure comprising data representing a plurality of states related to performance of the processing task, and data representing a plurality of selectable actions. The method obtains (306) requirements data describing requirements of the application and uses (308) the requirements data and the decision-making data structure to select an action to perform in relation to a layer of the neural network, and performs (310) the selected action to generate the task output for the application. The neural network comprises a convolutional neural network with encoder and decoder layers. The decision-making data ...

Video frame synthesis using tensor neural networks

A method for implementing video frame synthesis using a tensor neural network includes receiving input video data including one or more missing frames, converting the input video data into an input tensor, generating, through tensor completion based on the input tensor, output video data including one or more synthesized frames corresponding to the one or more missing frames by using a transform-based tensor neural network (TTNet) including a plurality of phases implementing a tensor iterative shrinkage thresholding algorithm (ISTA), and obtaining a loss function based on the output video data.

A system and method for monitoring system usage

A computer system for monitoring system usage under a service contract comprises a customer data processing module for receiving customer usage information and a machine learning module which comprises a trained data model that analyses the customer usage information and predicts future customer usage. An analysis module maps the predicted future customer usage onto the service contract and calculates the cost associated with the future usage and determines the excess cost of said future usage under the terms of service contract to allow the customer to amend usage or amend the contract in response to the determination of excess cost. The machine learning module may comprise a neural network and the trained data model may be trained using sample data which comprises historical usage information. This allows users to select the best service provider for the amount of usage they are predicted to use.

Audio generation methods and systems

A method of generating audio assets, comprising the steps of receiving an input multi‐layered audio asset S201 comprising a plurality of audio layers, generating an input multi‐channel image, wherein each channel of the input multi‐channel image comprises an input image representative of one of the audio layers S202, training a generative model on the input multi‐channel image and implementing the trained generative model to generate an output multi‐channel image, wherein each channel of the output multi‐channel image comprises an output image representative of an output audio layer S203, and generating an output multi‐layered audio asset based on a combination of output audio layers derived from the output images S204. The method may comprise arraigning the output layers with time delays between each of them. The input and output images may be spectrograms representing the audio data. The audio data may be taken from a video game and the generation of the output image may be influenced ...

Implementing a scatter function on a neural network accelerator

One or more vectors of indices are converted to sparse index tensors in a one-hot sparse format. An update tensor is generated, by applying the update values to one of the sparse index tensors (or a tensor derived from it). In some examples, an input data tensor is updated based on the update tensor. The update tensor may contain a plurality of input values to be updated. A sparse index tensor may be summed over one direction to calculate a reduced sparse index tensor, and element-wise multiplication performed. The neural network accelerator may use a delta function for clamp operation and a memory manipulation module to perform transpose, permute and repeat operations. This may be used in robotics or autonomous vehicles where Lidar produces a point cloud of object detection. The scatter operation may be used in steganography or watermarking.

HOLD GESTURE RECOGNITION USING MACHINE LEARNING

Embodiments are disclosed for hold gesture recognition using machine learning (ML). In an embodiment, a method comprises: receiving sensor signals indicative of a hand gesture made by a user, the sensor data obtained from at least one sensor of a wearable device worn by the user; generating a first embedding of first features extracted from the sensor signals; predicting a first part of a hold gesture based on a first ML gesture classifier and the first embedding; generating a second embedding of second features extracted from the sensor signals; predicting a second part of the hold gesture based on a second ML gesture classifier and the second embedding; predicting a hold gesture based at least in part on outputs of the first and second ML gesture classifiers and a prediction policy; and performing an action on the wearable device or other device based on the predicted hold gesture.

ANOMALY DETECTION METHOD, ANOMALY DETECTION DEVICE, AND RECORDING MEDIUM

An anomaly detection method by which a computer performs anomaly detection includes: obtaining first feature data outputted through N (N is an integer not less than 1) convolutional layers of a convolutional neural network configured as an encoder when an image is inputted to the convolutional neural network; obtaining second feature data outputted through M (M is an integer not less than 1, and M≠N) convolutional layers of the convolutional neural network and different in size from the first feature data; and performing anomaly detection on the image by using features indicated by the first feature data and the second feature data that are different in size.

Systems and methods for responding to predicted events in time-series data using synthetic profiles created by artificial intelligence models trained on non-homogenous time series-data

The systems and methods may use one or more artificial intelligence models that predict an effect of a predicted event on a current state of the system. For example, the model may predict how a rate of change in time-series data may be altered throughout the first time period based on the predicted event.

SYNAPTIC MEMORY AND MEMORY ARRAY USING FOWLER-NORDHEIM TIMERS

An analog memory device includes a first node and a second node. The first node includes a first floating gate, a second floating gate, and a capacitor. The first node first floating gate is connected to the first node second floating gate via the capacitor. The second node includes a first floating gate, a second floating gate, and a capacitor. The second node first floating gate is connected to the second node second floating gate via the capacitor. The second node is connected to the first node, and an analog state of the first node and an analog state of the second node continuously and synchronously decay with respect to time.

Deep Learning Model for Auto-Focusing Microscope Systems

A computing system receives, from an image sensor, at least two images of a specimen positioned on a specimen stage of a microscope system. The computing system provides the at least two images to an autofocus model for detecting at least one distances to a focal plane of the specimen. The computing system identifies, via the autofocus model, the at least one distance to the focal plane of the specimen. Based on the identifying, the computing system automatically adjusts a position of the specimen stage with respect to an objective lens of the microscope system.

Image processing neural networks with separable convolutional layers

A neural network system is configured to receive an input image and to generate a classification output for the input image. The neural network system includes: a separable convolution subnetwork comprising a plurality of separable convolutional neural network layers arranged in a stack one after the other, in which each separable convolutional neural network layer is configured to: separately apply both a depthwise convolution and a pointwise convolution during processing of an input to the separable convolutional neural network layer to generate a layer output.

Systems and Methods for Performing Image Enhancement using Neural Networks Implemented by Channel-Constrained Hardware Accelerators

Systems and methods for performing image enhancement using neural networks implemented by channel-constrained hardware accelerators in accordance with embodiments of the invention are described. One embodiment includes providing at least a portion of an input image to an input layer of a neural network implemented by a hardware accelerator, where the neural network has a spatial resolution and a number of channels and the input layer has initial spatial dimensions and an initial number of channels, performing an initial transformation operation based upon an input signal to produce an intermediate signal having reduced spatial dimensions and an increased number of channels, where: the reduced spatial dimensions are reduced relative to the initial spatial dimensions, and the increased number of channels is greater than the initial number of channels, processing the intermediate signal using the hardware accelerator based upon the parameters of the neural network to produce an initial output signal, performing a reverse transformation based upon the initial output signal to produce an output signal having increased spatial dimensions and a reduced number of channels, where: the increased spatial dimensions are increased relative to the reduced spatial dimensions; and the reduced number of channels is less than the increased number of channels, providing the output signal to an output layer of the neural network to generate at least a portion of an enhanced image, and outputting a final enhanced image using at least the at least a portion of an enhanced image.

Methods, systems, and apparatuses for user-understandable explainable learning models

Methods, systems, and apparatuses to build an explainable user output to receive input feature data by a neural network of multiple layers of an original classifier; determine a semantic function to label data samples with semantic categories; determine a semantic accuracy for each layer of the original classifier within the neural network; compare each layer based on results from the comparison of the semantic accuracy; designate a layer based on an amount of computed semantic accuracy; extend the designated layer by a category branch to the neural network to extract semantic data samples from the semantic content to train a set of new connections of an explainable classifier to compute a set of output explanations with an accuracy measure associated each output explanation for each semantic category of the plurality of semantic categories, and compare the accuracy measure for each output explanation to generate the output explanation in a user understandable format.

Camera Image Or Video Processing Pipelines With Neural Embedding

An image processing pipeline including a still or video camera includes a first portion of an image processing system arranged to use information derived at least in part from a neural embedding. A second portion of the image processing system can be used to modify at least one of an image capture setting, sensor processing, global post processing, local post processing, and portfolio post processing, based at least in part on neural embedding information.

Generating numeric embeddings of images

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating numeric embeddings of images. One of the methods includes obtaining training images; generating a plurality of triplets of training images; and training a neural network on each of the triplets to determine trained values of a plurality of parameters of the neural network, wherein training the neural network comprises, for each of the triplets: processing the anchor image in the triplet using the neural network to generate a numeric embedding of the anchor image; processing the positive image in the triplet using the neural network to generate a numeric embedding of the positive image; processing the negative image in the triplet using the neural network to generate a numeric embedding of the negative image; computing a triplet loss; and adjusting the current values of the parameters of the neural network using the triplet loss.

Systems and Methods of Connected Driving Based on Dynamic Contextual Factors

Systems including one or more sensors, coupled to a vehicle, may detect sensor information and provide the sensor information to another computing device for processing. A system includes one or more sensors, coupled to a vehicle and configured to detect sensor information, and a computing device configured to communicate with one or more mobile sensors to receive the mobile sensor information, communicate with the one or more sensors to receive the sensor information, and analyze the sensor information and the mobile sensor information to identify one or more risk factors.

Method for improving performance of a trained machine learning model

A method for improving performance of a trained machine learning model includes adding a second classifier with a second objective function to a first classifier with a first objective function. Rather than minimizing a function of errors for the first classifier, the second objective function is used to directly reduce the number errors of the first classifier.

System and method for distributive training and weight distribution in a neural network

A system for distributive training and weight distribution in a neural network. The system includes a training facility having a training neural network that detects and classifies objects in training images so as to train weights of nodes in the training neural network, and a plurality of object detection and classification units each including an image source that provides image frames, and at least one classification and prediction neural network that identifies, classifies and indicates relative velocity of objects in the image frames. Each unit transmits its image frames to the training facility so that the training neural network further trains the weights of the nodes in the training neural network, and the trained neural network weights are transmitted from the training facility to each of the object detection and classification units so as to train weights of nodes in the at least one classification and prediction neural network.

Convolutional Neural Network Processing Method and Device

A convolutional neural network (CNN) processing method and a CNN processing device are provided. The CNN processing method includes: acquiring an intensity type of a first module in a CNN model; in response of determining that the intensity type of the first module is a computation-intensity type, deploying the first module with an application specific integrated circuit (ASIC), where the first module occupies multiple arithmetic unit resources of the ASIC; acquiring the multiple arithmetic unit resources of the ASIC which are occupied by the first module; obtaining a first resource merging module by merging identical arithmetic unit resources among the multiple arithmetic unit resources of the ASIC which are occupied by the first module; and operating the first resource merging module by the ASIC.

Neural network pruning method and system via layerwise analysis

Embodiments disclosed herein allowed neural networks to be pruned. The inputs and outputs generated by a reference neural network are used to prune the reference neural network. The pruned neural network may have a subset of the weights that are in the reference neural network.

Speech recognition using convolutional neural networks

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing speech recognition by generating a neural network output from an audio data input sequence, where the neural network output characterizes words spoken in the audio data input sequence. One of the methods includes, for each of the audio data inputs, providing a current audio data input sequence that comprises the audio data input and the audio data inputs preceding the audio data input in the audio data input sequence to a convolutional subnetwork comprising a plurality of dilated convolutional neural network layers, wherein the convolutional subnetwork is configured to, for each of the plurality of audio data inputs: receive the current audio data input sequence for the audio data input, and process the current audio data input sequence to generate an alternative representation for the audio data input.

Automated generation of initial stimulation profile for sexual stimulation devices

A system and method for automated generation of control signals for sexual stimulation devices from usage history and other data. The system and method involve analyzing historical usage and other data for a user for a device or devices, processing the data through machine learning algorithms, and generating new or recombined patterns of stimulation based on the outputs from the machine learning algorithms. The resulting automated control signals represent partially or fully customized stimulation for a given user which evolve over time as the user continues to use the device or devices.

Using hardware-accelerated instructions

A computer-implemented method of implementing a computation using a hardware-accelerated instruction of a processor system by solving a constraint satisfaction problem. A solution to the constraint satisfaction problem represents a possible invocation of the hardware-accelerated instruction in the computation. The constraint satisfaction problem assigns nodes of a data flow graph of the computation to nodes of a data flow graph of the instruction. The constraint satisfaction problem comprises constraints enforcing that the assigned nodes of the computation data flow graph have equivalent data flow to the instruction data flow graph, and constraints restricting which nodes of the computation data flow graph can be assigned to the inputs of the hardware-accelerated instruction, with restrictions being imposed by the hardware-accelerated instruction and/or its programming interface.

Artificial intelligence-based legal document analysis system and method

Disclosed are an artificial intelligence-based legal document analysis system and method. The present invention can provide relevant laws and detailed explanation by analyzing the legal risk in a legal document having a structure such as legal clauses, terms and conditions and contracts by automatically comprehending the meaning by means of an artificial intelligence technology, and perceiving omissions and erroneous risk elements in the contract.

Method and system for training a neural network

A method for training a neural network for detecting a plurality of classes of object within a sample comprises providing a training data set comprising a plurality of samples, each annotated according to whether the samples include labelled objects of interest. In a first type of samples, all objects of interest are labelled according to their class and comprise a foreground of said samples, the remainder of the samples comprising background. In a second type of samples, some objects of interest are labelled in a foreground and their background may comprise unlabelled objects. A third type of samples comprise only background comprising no objects of interest. Negative mining is only performed on the results of processing the first and third types of samples.

Method for generating learning model and program

There is provided a method for generating a learning model, the method including: acquiring an endoscopic image captured by an endoscope and manipulation information regarding a manipulation of an endoscope operator in each stage of operation of the endoscope by the endoscope operator operating the endoscope; and generating a learning model learned so as to output the manipulation information of a next stage in a case where the endoscopic image and the manipulation information are input, based on training data including the acquired endoscopic image and manipulation information, and the manipulation information of the next stage.

Systems and Methods for Human Activity Recognition Using Analog Neuromorphic Computing Hardware

Systems, methods, and devices are provided for human activity recognition. An example device includes an integrated circuit for human activity recognition. The integrated circuit includes an analog network of analog components configured to implement a trained neural network model (e.g., an autoencoder) that is trained to generate a plurality of descriptors for a plurality of predefined human activities based on a plurality of features extracted from a plurality of electrical signals from one or more sensors. The device also includes one or more digital components configured to classify human activity (e.g., using a classifier, such as K-Nearest Neighbor) as one of the plurality of predefined human activities according to the plurality of descriptors generated by the integrated circuit. In some implementations, the device further includes the one or more sensors configured to collect the plurality of electrical signals during the human activity.

Machine learned anomaly detection

A computer-implemented method and system for training an anomaly detector to distinguish outlier data from inlier data on which the anomaly detector is trained. The anomaly detector comprises a set of learnable data transformations and a learnable feature extractor. The set of learnable data transformations and the learnable feature extractor are jointly trained based on a trained objective, which training objective comprises a function serving as anomaly scoring function which may also be used at test time to determine the anomaly score of test data samples. Evaluation results show that the anomaly detector is well-applicable to detect anomalies in non-image data, e.g., in data timeseries and in tabular data, and straightforward to apply at test time.

Image processing device and operating method thereof

An image processing device, including a memory configured to store one or more instructions; and a processor configured to execute the one or more instructions stored in the memory to: obtain kernel coefficient information corresponding to each pixel of a plurality of pixels included in a first image, using a convolution neural network including one or more convolution layers, generate a spatially variant kernel including a kernel corresponding to the each pixel, based on a gradient kernel set including a plurality of gradient kernels corresponding to one or more gradient characteristics of the plurality of pixels, and the kernel coefficient information, and generate a second image, by applying the kernel included in the spatially variant kernel to a region centered on the each pixel, and filtering the first image.

Offensive chat filtering using machine learning models

A server system is provided that includes one or more processors configured to execute a platform for an online multi-user chat service that communicates with a plurality of client devices of users of the online multi-user chat service that exchanges user chat data between the plurality of client devices. The one or more processors are configured to execute a user chat filtering program that performs filter actions for user chat data exchanged on the platform for the online multi-user chat service. The user chat filtering program includes a plurality of trained machine learning models and a filter decision service that determines a filter action to be performed for target portions of user chat data based on output of the plurality of trained machine learning models for those target portions of user chat data.

Target cell statistical method, apparatus, and system

A target cell statistical method, apparatus and system are provided. A cell image of a blood specimen is acquired by a cell image analysis apparatus. The blood specimen is derived from a blood sample to be tested. A number of target cells and a number of reference cells in the cell image are automatically identified by the cell image analysis apparatus. A number of reference cells in the blood sample to be tested is acquired by the cell image analysis apparatus, and a number of target cells in the blood sample to be tested is calculated by the cell image analysis apparatus, based on the number of target cells and the number of reference cells in the cell image and the number of reference cells in the blood sample to be tested.

Shape-biased image classification using deep convolutional networks

A system for analyzing images includes a processing device includes a receiving module configured to receive an image, and an analysis module configured to apply the received image to a machine learning network and classify one or more features in the received image, the machine learning network configured to propagate image data through a plurality of convolutional layers, each convolutional layer of the plurality of convolutional layers including a plurality of filter channels, the machine learning network including a bottleneck layer configured to recognize an image feature based on a shape of an image component, The system also includes an output module configured to output characterization data that includes a classification of the one or more features.

Convolutional arithmetic processing device and convolutional arithmetic processing system

A convolutional arithmetic processing device includes a convolutional arithmetic processor and a storage device. The convolutional arithmetic processor performs a first convolutional arithmetic process of a convolutional neural network on numerical values of a first three-dimensional array, using a type of kernel formed of numerical values of a second three-dimensional array, where a number of the type is represented by a second numerical value with a stride represented by a third numerical value in a first direction and a stride represented by a fourth numerical value in a second direction. The storage device stores at least part of the numerical values of the first three-dimensional array.

Digital quality control using computer visioning with deep learning

Implementations include receiving sample data, the sample data being generated as digital data representative of a sample of the product, providing a set of features by processing the sample data through multiple layers of a residual network, a first layer of the residual network identifying one or more features of the sample data, and a second layer of the residual network receiving the one or more features of the first layer, and identifying one or more additional features, processing the set of features using a CNN to identify a set of regions, and at least one object in a region of the set of regions, and determine a type of the at least one object, and selectively issuing an alert at least partially based on the type of the at least one object, the alert indicating contamination within the sample of the product.

Neural Network Architecture Using Control Logic Determining Convolution Operation Sequence

Hardware for implementing a Deep Neural Network (DNN) having a convolution layer, the hardware comprising a plurality of convolution engines each operable to perform a convolution operation by applying a filter to a data window, each filter comprising a set of weights for combination with respective data values of a data window, and each of the plurality of convolution engines comprising: multiplication logic operable to combine a weight of a filter with a respective data value of a data window; control logic configured to: receive configuration information identifying a set of filters for operation on a set of data windows at the plurality of convolution engines; determine, using the configuration information, a sequence of convolution operations for evaluation at the multiplication logic; in accordance with the determined sequence of convolution operations, request weights and data values for at least partially applying a filter to a data window; and cause the multiplication logic to combine the weights with their respective data values; and accumulation logic configured to accumulate the results of a plurality of combinations performed by the multiplication logic so as to form an output for a convolution operation of the determined sequence.

Error correction in network packets

Systems and methods for error correction in network packets are provided. An example method includes receiving a network packet via a communication channel, the network packet including a content and an error-detecting code associated with the content, determining, based on the error-detecting code, that the network packet is corrupted, selecting a pre-determined number of positions of bits in the content of the network packet, changing values of the bits in the selected positions to a bit value combination selected from all possible bit value combinations in the selected positions to modify the content and calculating a further error-detecting code of the modified content until the further error-detecting code of the modified payload matches the error-detecting code received via the communication channel or all possible bit combinations have been selected, and if the further error-detecting code does not match the error-detecting code, requesting for retransmission of the network packet.

Method and system for automatic identification of primary manufacturing process from three-dimensional model of product

The invention relates to method and system for automatic identification of a primary manufacturing process (PMP) from a three-dimensional (3D) model of a product. The method includes generating a plurality of images corresponding to a plurality of views of the product based on the 3D model of the product; determining a plurality of confidence score vectors, based on the plurality of images, using a first Artificial Neural Network (ANN) model; determining an aggregate confidence score vector, representing a pre-defined PMP category with maximum frequency, based on the plurality of confidence score vectors; extracting a set of manufacturing parameters associated with the product, based on the 3D model of the product; and identifying the PMP based on the aggregate confidence score vector and the set of manufacturing parameters, using a second ANN model.

Systems and methods for self-learned label refinement for improving monocular object detection

Described are systems and methods for self-learned label refinement of a training set. In on example, a system includes a processor and a memory having a training set generation module that causes the processor to train a model using an image as an input to the model and 2D bounding based on 3D bounding boxes as ground truths, select a first subset from predicted 2D bounding boxes previously outputted by the model, retrain the model using the image as the input and the first subset as ground truths, select a second set of predicted 2D bounding boxes previously outputted by the model, and generate the training set by selecting the 3D bounding boxes from a master set of 3D bounding boxes that have corresponding 2D bounding boxes that form the second subset.

Evaluation process for a multi-task network

System, methods, and other embodiments described herein relate to evaluating a perception network in relation to the accuracy of depth estimates and object detections. In one embodiment, a method includes segmenting range data associated with an image according to bounding boxes of objects identified in the image to produce masked data. The method includes comparing the masked data with corresponding depth estimates in the depth map according to an evaluation mask that correlates the depth estimates with the depth map. The method includes providing a metric that quantifies the comparing to assess a network that generated the depth map and the bounding boxes.

Multi-task self-training for learning general representations

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training an image representation neural network.

Data volume sculptor for deep learning acceleration

A device include on-board memory, an applications processor, a digital signal processor (DSP) cluster, a configurable accelerator framework (CAF), and at least one communication bus architecture. The communication bus communicatively couples the applications processor, the DSP cluster, and the CAF to the on-board memory. The CAF includes a reconfigurable stream switch and data volume sculpting circuitry, which has an input and an output coupled to the reconfigurable stream switch. The data volume sculpting circuitry receives a series of frames, each frame formed as a two dimensional (2D) data structure, and determines a first dimension and a second dimension of each frame of the series of frames. Based on the first and second dimensions, the data volume sculpting circuitry determines for each frame a position and a size of a region-of-interest to be extracted from the respective frame, and extracts from each frame, data in the frame that is within the region-of-interest.

System and method for completing continual multi-agent trajectory forecasting

A system and method for completing continual multi-agent trajectory forecasting with a graph-based conditional generative memory system that include receiving data associated with a surrounding location of an ego agent and inputting the data associated with the surrounding location of the ego agent to at least one episodic memory buffer and processing scene graphs associated with the surrounding location of the ego agent that are associated with the plurality of time steps. The system and method additionally include aggregating the data associated with the surrounding location of the ego agent associated with the plurality of time steps into mixed data and training a generative memory and a predictor with the mixed data. The system and method further include predicting future trajectories associated with traffic agents that are located within the surrounding location of the ego agent based on the training of the generative memory and the predictor.

Method and system for product knowledge fusion

A method and system for product knowledge fusion are discloses. The method includes following steps: acquiring original data of a product; performing knowledge extraction on the original data of the product to obtain entities, attributes and semantic relationships related to the product; building an entity information knowledge base according to the entities, attributes and semantic relationships related to the products; fusing the semantic relationships and attributes with the entities and matching the entities by adopting a text matching model to obtain original data of the product corresponding to matched entity information; and establishing a knowledge graph of the product according to the matched entity information. The method and system standardize multi-source heterogeneous data with a knowledge fusion method, thus effectively reducing polysemy and unclear references of knowledge caused by different data structures and sources.

Methods and devices for structured pruning for automatic speech recognition

Methods and apparatuses for automatic speech recognition are provided. The method includes: generating a weight matrix for a layer of a plurality of layers in a neural network; dividing the weight matrix into a plurality of blocks, each block including a plurality of weights; selecting a set of blocks from the plurality of blocks for block-wise pruning by minimizing a cost function subject to a pre-determined block-wise constraint; and generating a block-wise pruned weight matrix by setting one or more weights in the set of blocks to zero. The weight matrix includes a set of weights associated with the layer, the plurality of layers includes a first layer receiving a first input associated with one or more audio feature sequences, and the plurality of layers are executed on one or more processors.

System and method for automatically switching a vehicle to follow in a vehicle's autonomous driving mode

Systems and methods for implementing one or more autonomous features for autonomous and semi-autonomous control of one or more vehicles are provided. More specifically, image data may be obtained from an image acquisition device and processed utilizing one or more machine learning models to identify, track, and extract one or more features of the image utilized in decision making processes for providing steering angle and/or acceleration/deceleration input to one or more vehicle controllers. In some instances, techniques may be employed such that the autonomous and semi-autonomous control of a vehicle may change between vehicle follow and lane follow modes. In some instances, at least a portion of the machine learning model may be updated based on one or more conditions.

Imaging device and electronic device

The present technology relates to an imaging device and an electronic device that enable construction of an imaging device that outputs information required by a user with a small size. A single-chip imaging device includes: an imaging unit in which a plurality of pixels is arranged two-dimensionally and that captures an image; a signal processing unit that performs signal processing using a captured image output from the imaging unit; an output I/F that outputs a signal processing result of the signal processing and the captured image to an outside; and an output control unit that performs output control of selectively outputting the signal processing result of the signal processing and the captured image from the output I/F to the outside. The present technology can be applied to, for example, an imaging device that captures an image.

Machine vision control system for precision agriculture

An illustrative control system for an precision agricultural implement includes a controller having a convolutional neural network, a machine vision module, a plurality of sensors, and a plurality of actuators in communication with the controller, the plurality of actuators including a plurality of agricultural tool actuators.

Systolic neural network engine capable of forward propagation

A method of computer processing is disclosed comprising receiving a data packet at a processing node of a neural network, performing a calculation of the data packet at the processing node to create a processed data packet, attaching a tag to the processed data packet, transmitting the processed data packet from the processing node to a receiving node during a systolic pulse, receiving the processed data packet at the receiving node, performing a clockwise convolution on the processed data packet and a counter clockwise convolution on the processed data packet, performing an adding function and backpropagating results of the performed sigmoid function to each of the processing nodes that originally processed the data packet.

Hierarchical weight preprocessing for neural network accelerator

A system and method for weight preprocessing. In some embodiments, the method includes performing intra-tile preprocessing of a first weight tensor to form a first pre-processed weight tensor, and performing inter-tile preprocessing of the first pre-processed weight tensor, to form a second pre-processed weight tensor. The intra-tile preprocessing may include moving a first element of a first weight tile of the first weight tensor by one position, within the first weight tile, in a lookahead direction or in a lookaside direction. The inter-tile preprocessing may include moving a first row of a weight tile of the first pre-processed weight tensor by one position in a lookahead direction or by one position in a lookaside direction.

Recurrent environment predictors

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for environment simulation. In one aspect, a system comprises a recurrent neural network configured to, at each of a plurality of time steps, receive a preceding action for a preceding time step, update a preceding initial hidden state of the recurrent neural network from the preceding time step using the preceding action, update a preceding cell state of the recurrent neural network from the preceding time step using at least the initial hidden state for the time step, and determine a final hidden state for the time step using the cell state for the time step. The system further comprises a decoder neural network configured to receive the final hidden state for the time step and process the final hidden state to generate a predicted observation characterizing a predicted state of the environment at the time step.

Robotic tattooing systems and related technologies

An automatic tattoo apparatus can be used to robotically apply tattoos. A customer can shop on an online tattoo marketplace to select designs created by various artists located anywhere. The online tattoo marketplace can process and manage payments, artist and/or customer profiles, bookings, tattoo design uploads, browsing and design selection, design changes, and/or perform other actions. The automatic tattoo device can apply tattoos precisely, quickly, and may with reduced pain. The tattoo apparatus can apply a wide range of different types of tattoos, including but not limited to micro tattoos, dotwork, blackwork tattoos, realism tattoos, fine-line tattoos, etc.

Automatic image selection for online product catalogs

Disclosed are systems, methods, and non-transitory computer-readable media for automatic image selection for online product catalogs. An image selection system gathers feature data for images of an item included in listings posted to an online marketplace. The image selection system uses the feature data as input in a machine learning model to determine probability scores indicating an estimated probability that each image is suitable to represent the item. The machine learning model is trained based on a set of training images of the item that have been labeled to indicate whether they are suitable to represent the image. The image selection system compares the probability scores and selects an image to represent the item as a stock image based on the comparison.

Imaging device and electronic device

The present technology relates to an imaging device and an electronic device that enable construction of an imaging device that outputs information required by a user with a small size. A single-chip imaging device includes: an imaging unit in which a plurality of pixels is arranged two-dimensionally and that captures an image; a signal processing unit that performs signal processing using a captured image output from the imaging unit; an output I/F that outputs a signal processing result of the signal processing and the captured image to an outside; and an output control unit that performs output control of selectively outputting the signal processing result of the signal processing and the captured image from the output I/F to the outside. The present technology can be applied to, for example, an imaging device that captures an image.

Detecting abnormalities in ecg signals

A method of detecting abnormalities in ECG signals by providing an ECG signal to a neural network, performing a first series of convolution operations to a first subset of layers and in a final layer, and determining a plurality of preliminary classification estimates, each preliminary classification estimate corresponding with a time segment of the ECG signal. Furthermore, determining input data for a second subset of layers of the neural network by concatenating the preliminary classification with the output of a layer of the first subset of layers that precedes the final layer of the first subset of layers. Within the second subset of layers of the neural network, performing a second series of convolution operations. In a final layer of the second subset, determining plurality of final classification estimates, each final classification estimate corresponding with a time segment of the ECG signal.

Predicting topics of potential relevance based on retrieved/created digital media files

Implementations are described herein for leveraging digital media files retrieved and/or created by users to predict/determine topics of potential relevance to the users. In various implementations, digital media file(s) created and/or retrieved by a user with a client device may be applied as input across trained machine learning model(s), which in some cases are local to the client device, to generate output that indicates object(s) detected in the digital media file(s). Data indicative of the indicated object(s) may be provided to a remote computing system without providing the digital media file(s) themselves. In some implementations, information associated with the indicated object(s) may be retrieved and proactively output to the user. In some implementations, a frequency at which objects occur across a corpus of digital media files may be considered when determining a likelihood that a detected object is potentially relevant to a user.

Adaptive high-precision compression method and system based on convolutional neural network model

The present disclosure discloses an adaptive high-precision compression method and system based on a convolutional neural network model, and belongs to the fields of artificial intelligence, computer vision, and image processing. According to the method of the present disclosure, coarse-grained pruning is performed on a neural network model by using a differential evolution algorithm first, and the coarse-grained space is quickly searched through an entropy importance criterion and an objective function with good guidance to obtain a near-optimal neural network structure. Then fine-grained search space is built on the basis of an optimal individual obtained from the coarse-grained search, and fine-grained pruning is performed on the neural network model by a differential evolution algorithm to obtain a network model with an optimal structure. Finally, the performance of the optimal model is restored by using a multi-teacher multi-step knowledge distillation network to reach the precision of an original model.

Biofeedback-based control of sexual stimulation devices

A system and method for biofeedback-based control of sexual stimulation devices involving receiving biometric data, analyzing the biometric data to detect changes in the physiology of a person, and generating control signals based on the changes. In some embodiments, the analyses of the biometric data are performed by machine learning algorithms which may be trained on associations between biometric data of a user, indications of the user's state of arousal, and the state of operation of a sexual stimulation device. In some embodiments, machine learning algorithms are used to make the associations. In some embodiments, biofeedback-based controls may be incorporated into systems of controls comprising thought-based controls and/or voice-based controls.

Method for determining degree of response to physical activity

The present invention discloses a method for determining a degree of response to a physical activity. Acquire a physical activity signal measured by a sensing unit in the physical activity. Determine first data of a first physical activity feature set based on the physical activity signal. Determine a recognition of the degree of response to the physical activity based on the first data of the first physical activity feature set by a mathematical model describing a relationship between the first physical activity feature set and the degree of response to a physical activity. A portion of a first mechanism of the mathematical model adopts at least one portion of a second mechanism of a first neural network model associated with the second physical activity feature set.

Deep learning accelerator models and hardware

A first deep learning accelerator (DLA) model can be executed using a first subset of a plurality of DLA cores of a DLA chip. A second DLA model can be executed using a second subset of the plurality of DLA cores of the DLA chip. The first subset can include a first quantity of the plurality of DLA cores. The second subset can include a second quantity of the plurality of DLA cores that is different than the first quantity of the plurality of DLA cores.

Automated inventory management method and system thereof

An automated inventory management method is provided. A historical sale record is received, and a future sale of an item is predicted based on the historical sale record to obtain a simulation result of an expected sale state of the item in the next sale cycle. According to the historical sale record of full categories of items and the simulation result of the item in the next sale cycle, an initial weight of the pre-training model is trained and used as a weight of an inventory decision module, and a purchase order that meets the expected sale record of the item in the next sale cycle is automatically generated. A reward feedback is calculated according to a current sale record and an inventory volume of the item and a purchase order of the previous sale cycle and input into the inventory decision module to order the item.

Optimal power flow acquiring method for regional distribution network of small hydropower groups based on deep learning

Disclosed is an optimal power flow acquiring method for regional distribution network of small hydropower groups based on deep learning, which specifically includes the following steps: generating required data sets by adopting continuous power flow and power flow equation calculation methods; the data set is randomly divided into training data (80 percent) and test data (20 percent); training the built convolutional neural network model with training data to learn the mapping relationship between load and generator output power; inputting test data, and directly obtaining PG and QG from the trained convolutional neural network; and solving residual variables Vi and θi with traditional power flow solver. The application can accelerate the solving speed of the optimal power flow problem with higher prediction accuracy.

Enhancement of weak signal for machine training neural network representing a solid-state detector

A physics-based network model is trained to learn weights such as trapping, detrapping, and/or transport of holes and/or electrons, as well as voltage distribution on a voxel-by-voxel basis throughout a solid-state detector model. The physics-based network may be used to estimate material property variation throughout the voxels. Anode and cathode signals as well as the voltage distribution are relatively strong signals compared to the weaker electron and hole signals. The relatively weaker signals may be limited in range across voxels. In order to expand the range or magnify the effect, the loss function used in training the physics-based neural network may use a weighted combination where the weaker signals are weighted more heavily than stronger signals without substantially reducing the influence of the stronger signals. This improves the inference, resulting in improvement of the accuracy and range of the trained physics-based model.

Blindness assist glasses

An eyewear device with camera-based compensation that improves the user experience for user's having partial blindness or complete blindness. The camera-based compensation determines objects, converts determined objects to text, and then converts the text to audio that is indicative of the objects and that is perceptible to the eyewear user. The camera-based compensation may use a region-based convolutional neural network (RCNN) to generate a feature map including text that is indicative of objects in images captured by a camera. Relevant text of the feature map is then processed through a text to speech algorithm featuring a natural language processor to generate audio indicative of the objects in the processed images.

Data processing method and circuit based on convolution computation

A data processing method and circuit based on convolution computation are provided. In the data processing method, a shared memory structure is provided, convolution computation of data in batches or duplicated data is provided, an allocation mechanism for storing data into multiple memories is provided, and a signed padding mechanism is provided. Therefore, a flexible and efficient convolution computation mechanism and structure are provided.

Method and system for anamoly detection in the banking system with graph neural networks (gnns)

A method and system for anomaly detection in the banking system with graph network of a plurality of interconnected gateways. The system continuously monitors a plurality of gateways, data flows related to and executed at a first gateway of the plurality of gateways, the gateway data flows including at least one or more of gateways in a network.

Methods and systems for computing an output of a neural network layer

Systems and methods for computing a neural network layer of a neural network are described. A squared Euclidean distance is computed between the input vector and the weight vector of the neural network layer, replacing computation of the inner product. Methods for quantization of the squared Euclidean computation are also described. Methods for training the neural network using homotopy training are also described.

Deep Learning Models for Region-of-Interest Determination

A method of determining a region of interest in an image of tissue of an individual by an apparatus including processing circuitry may include executing, by the processing circuitry, instructions that cause the apparatus to partition an image of tissue of an individual into a set of areas, identify a tissue type of each area of the image, and apply a classifier to the image to determine a region of interest, the classifier being configured to determine regions of interest based on the tissue types of the set of areas of the image.

System, devices and/or processes for adapting neural network processing devices

Example methods, apparatuses, and/or articles of manufacture are disclosed that may be implemented, in whole or in part, using one or more computing devices to adapt a computing device to classify physical features in a deployment environment. In a particular implementation, computing resources may be selectively de-allocated from at least one of one or more elements of a computing architecture based, at least in part, on assessed impacts to the one or more elements of the computing architecture.

Prediction method and system of low blood pressure

A prediction method and system of a low blood pressure is provided, including: obtaining a plurality of feature sequence values; selecting two of the feature sequence values from the feature sequence values according to a time ratio relationship; calculating a relation coefficient according to the selected two feature sequence values by a weighting process; repeating to select the new feature sequence values and the corresponding relation coefficient and to assign the new feature sequence values and the relation coefficient into the input group until the feature sequence values conforming to the time ratio relationship are traversed; and obtaining a training result by substituting the input group into a low blood pressure prediction model.

Edge computing device for controlling electromechanical system or electronic device with local and remote task distribution control

A computing device, including a processor configured to receive sensor data from a control device. The control device may include a control processor configured to execute control instructions to control an actuator of a target electromechanical system and may further include one or more sensors. The processor may identify a first subset of the sensor data and a second subset of the sensor data. The processor may generate first control instructions based on the first subset and transmit the first control instructions to the control processor of the control device. The processor may transmit the second subset to a remote computing device. In response to transmitting the second subset to the remote computing device, the processor may receive a remote processing result from the remote computing device. The processor may generate second control instructions from the remote processing result and transmit the second control instructions to the control processor.

Optical computing apparatus and system, and computing method

An optical computing apparatus and system and a computing method are provided. The optical computing apparatus includes a linear operation module, a first delay module, and a coupler. The linear operation module can modulate, based on received electrical signals, optical signals input to the linear operation module; the first delay module may adjust a delay of optical signals output by the linear operation module; and after the first delay module adjusts the delay of the optical signals output by the linear operation module, the coupler may combine a plurality of groups of optical signals successively output by the linear operation module, to output one group of optical signals used to indicate a computing result that is obtained after a multiply-add operation is performed on one group of data and weights.

Item condition prediction operations and interfaces in an item listing system

Various methods and systems for providing predicted item conditions for items in an item listing system. A predicted item condition may indicate a calculated estimate of a descriptive state of the item based on item transaction features. Operationally, item transaction features of an item—associated with an item listing interface—are accessed at the item listing system. The item transaction features are communicated to an item condition prediction machine learning model of the item listing system. The item condition machine learning model is trained on historical item transactions comprising item condition features of historical item transactions, the historical item transactions are previous item transactions associated with the item listing system. Based on the item transaction features of the item, the item condition machine learning model is caused to generate a predicted item condition. The predicted item condition is communicated as a recommended item condition or required item condition.

Training apparatus, training method, and medium

A training apparatus is provided. The training apparatus acquires a mosaic image, generates a demosaic image by subjecting the mosaic image to a demosaicing process in which a neural network is used, and detects a low-image-quality portion in the demosaic image as a detected region. The training apparatus acquires a training image including a region having a hue similar to a hue of the detected region, and incrementally trains the neural network using the training image.

Image dehazing method and system based on cyclegan

Disclosed are an image dehazing method and system based on CycleGAN. The method comprises: acquiring a to-be-processed hazy image; and inputting the image into a pre-trained densely connected CycleGAN, and outputting a clear image. The densely connected CycleGAN comprises a generator, the generator comprises an encoder, a converter and a decoder, the encoder comprises a densely connected layer for extracting features of an input image, the converter comprises a transition layer for combining the features extracted at the encoder stage, the decoder comprises a densely connected layer and a scaled convolutional neural network layer, the densely connected layer is used for restoring original features of the image, and the scaled convolutional neural network layer is used for removing a checkerboard effect of the restored original features to obtain a finally output clear image.

Model update method and related apparatus

A method includes: performing a plurality of times of first clustering processing on a plurality of training images based on a plurality of target features to obtain a plurality of first clustering results, where each of the plurality of first clustering results corresponds to one silhouette coefficient, and the silhouette coefficient indicates cluster quality; determining a first target clustering result based on the silhouette coefficients, where the first target clustering result includes M clustering categories; and performing second clustering processing on the plurality of training images based on the plurality of target features to obtain a second clustering result, where a quantity of clustering categories included in the second clustering result is M.

Multi-view deep neural network for lidar perception

A deep neural network(s) (DNN) may be used to detect objects from sensor data of a three dimensional (3D) environment. For example, a multi-view perception DNN may include multiple constituent DNNs or stages chained together that sequentially process different views of the 3D environment. An example DNN may include a first stage that performs class segmentation in a first view (e.g., perspective view) and a second stage that performs class segmentation and/or regresses instance geometry in a second view (e.g., top-down). The DNN outputs may be processed to generate 2D and/or 3D bounding boxes and class labels for detected objects in the 3D environment. As such, the techniques described herein may be used to detect and classify animate objects and/or parts of an environment, and these detections and classifications may be provided to an autonomous vehicle drive stack to enable safe planning and control of the autonomous vehicle.

神经网络模型处理方法、装置、计算机设备及存储介质

本申请实施例公开了一种神经网络处理方法、装置、计算机设备及存储介质，把一个算子拆分成多个规模更小的子算子，这样可以直接调用单核架构下的计算库，充分利用了多核处理器的硬件资源，从而可以避免重现实现的额外工作量。

用于异常检测和/或预测性维护的计算机实现的方法、计算机程序产品以及系统

提供了一种用于异常检测和/或预测性维护的计算机实现的方法和相应的系统。该方法包括：接收新观测值，该新观测值表征实体的至少一个参数；将新观测值输入到深度神经网络(100)，该深度神经网络具有多个隐藏层并且使用训练数据集进行训练，该训练数据集包括可能的观测值；获得通过将接收到的新观测值输入深度神经网络而从深度神经网络的多个隐藏层中的至少一个输出的第二中间输出值集；使用存储在存储介质中的潜变量模型来将第二中间输出值集映射到第二投影值集；基于潜变量模型和第二投影值集来确定接收到的新观测值相对于训练数据集是否为异常值，通过深度神经网络来计算针对新观测值的预测；以及基于预测和新观测值是否为异常值的确定来确定指示实体中的至少一个异常的出现的结果。通过以下操作来构建存储在存储介质中的潜变量模型：获得从深度神经网络的多个隐藏层中的所述一个输出的第一中间输出值集，通过输入训练数据集的至少一部分中包括的可能观测值中的不同一者来获得第一中间输出值集中的每一个；以及使用第一中间输出值集来构建潜变量模型，潜变量模型提供第一中间输出值集到潜变量模型的子空间中的第一投影值集的映射，该子空间具有低于输出值集的维度的维度。

心电数据的分类方法及分类系统

本发明实施例提供一种心电数据的分类方法及分类系统，涉及医疗信息技术领域，可以提高处理心电数据的准确度和泛化能力。一种心电数据的处理方法，包括：将心电数据分割成多个心搏数据；对多个心搏数据进行卷积和池化，得到第一特征向量；获取第二特征向量，所述第二特征向量包括多个所述心搏数据的频域特征数据和时域特征数据；对所述第一特征向量和所述第二特征向量融合，生成融合特征向量；根据所述融合特征向量，得到所述心电数据的分类信息。

串流开关

本公开涉及串流开关。实施例涉及形成在集成电路中的可重新配置的串流开关。串流开关包括多个输出端口、多个输入端口和多个选择电路。输出端口均具有输出端口架构组成，并且每个输出端口被布置为单向传递输出数据和输出控制信息。输入端口均具有输入端口架构组成，并且每个输入端口被布置为单向接收第一输入数据和第一输入控制信息。选择电路中的每一个被耦合到输出端口的相关联的一个。每个选择电路还被耦合到所有输入端口，使得每个选择电路被布置为在任何给定时间将其相关联的输出端口可重新配置地耦合到不超过一个的输入端口。

Automatic wound boundary detecting method using artificial intelligence and 3d model generating method for wound

본 명세서는 정확하게 환부의 경계를 자동으로 인식할 수 있는 방법 및 인식된 환부 경계를 기준으로 3차원 환부 모델을 생성하는 방법을 개시한다. 본 명세서에 따른 환부 경계 자동 인식 방법은 인공지능에 기반한 환부 경계 자동 인식 방법으로서, 인식하고자 하는 환부를 RGB-D 카메라로 여러 프레임을 촬영하고, 이미지 내 측정 정보를 분리, 학습을 위한 이미지 데이터 증폭을 거쳐서 인공신경망을 거칠 수 있다. 인공신경망을 거친 자료에 대한 경계 인식 후처리를 통해 2D 이미지를 3D 모델로 매칭하여 3D 모델을 생성하는 단계;를 포함할 수 있다.

User equipment, base station, channel estimation and feedback system of user equipment and base station

本公开提供了一种无线通信中的用户设备、基站、用户设备和基站的联合训练设备、用户设备和基站的联合信道估计和反馈系统、用户设备执行的反馈信道状态信息生成方法、基站执行的信道矩阵生成方法、用户设备和基站的联合训练方法以及用于用户设备和基站的联合信道估计和反馈方法。通过由用户设备根据实际的导频信号生成反馈信道状态信息，在基站中引入更深层次的残差学习神经网络来根据反馈信道状态信息重建基站的信道矩阵。实现了即使在实际接受的导频信号为是非完整的低分辨率部分的情况下，基站也能重建完成的高分辨率信道矩阵。

Neural network processor

본 명세서는 낮은 비용으로 높은 처리율을 가진 연산 처리 장치를 개시한다. 본 명세서에 따른 연산 처리 장치는 뉴럴 네트워크의 프로세싱을 수행하기 위한 연산에 필요한 데이터를 메모리에서 읽어와 연산 유닛에 제공하는 페치 유닛을 포함하는 연산 처리 장치로서, 상기 페치 유닛은 상기 각 데이터 메모리 슬라이스에 저장된 데이터가 페치되는 페치 버퍼; 및 상기 페치 버퍼에 페치된 데이터에 각각의 데이터 메모리 슬라이스에 대응하는 노드 ID를 부여하고, 상기 노드 ID에 따라 상기 페치된 데이터를 출력하는 타이밍을 제어하는 인터페이스 컨트롤러;를 포함할 수 있다.

System on chip and mobile computing device

本公开涉及片上系统和移动计算设备。实施例针对实现深度卷积网络异构架构的片上系统(SoC)。SoC包括系统总线、耦合到系统总线的多个可寻址存储器阵列、耦合到系统总线的至少一个应用处理器核心以及耦合到系统总线的可配置的加速器框架。可配置的加速器框架是图像和深度卷积神经网络(DCNN)协同处理系统。SoC还包括耦合到系统总线的多个数字信号处理器(DSP)，其中多个DSP与可配置的加速器框架协调功能来执行DCNN。

Apparatus and method for estimating bio-information

생체정보 추정 장치가 개시된다. 일 실시예에 따르면 생체정보 추정 장치는 피검체로부터 맥파신호를 측정하는 맥파센서, 피검체와 맥파센서 사이에 작용하는 힘신호를 측정하는 힘센서, 및 맥파신호 및 힘신호를 기초로 제1 입력값, 제2 입력값 및 제3 입력값을 획득하고, 획득된 제1 입력값, 제2 입력값 및 제3 입력값을 제1 신경망 모델에 입력하여 특징 벡터를 추출하며, 추출된 특징 벡터를 제2 신경망 모델에 입력하여 생체정보를 획득하는 프로세서를 포함할 수 있다.

Object detection method and device based on reconfigurable network

本发明提供一种利用目标对象预测网络和目标对象集成网络，来学习适合诸如关键绩效指标(Key Performance Index)的用户要求的，基于CNN的对象检测器的参数的方法。所述CNN，根据所述关键绩效指标的分辨率或焦距发生变化而变化的对象的规模来进行重新设计。所述方法包括，学习装置使卷积层，对在图像上与第(k‑1)目标区域对应的第k处理图像进行卷积运算而输出第k特征图；以及使所述对象集成网络集成从FC层输出的第一至第n对象检测信息，并将参考所述集成对象检测信息和与其对应的GT来生成的损失反向传播。所述方法提高了2D边界框的准确度，从而可有效地执行于在多摄像机、环绕视图监控(Surround View Monitoring)等。

Systems and methods for training prediction system for depth perception

【課題】境界ボックスを使用して奥行き予測システムを訓練するシステム及び方法を提供すること。【解決手段】一実施形態では、本方法は、画像をセグメント化することにより、境界ボックスを超えた領域をマスクして、境界ボックスの内部の非マスク領域を識別することを含む。該方法は、また、非マスク領域内で画像の画素と関連付けられた重み付け点をグランドトゥルース奥行きと比較することからの奥行き損失を使用して、奥行きモデルを訓練することを含む。該方法は、また、物体検出のために奥行きモデルを提供することを含む。【選択図】図５

Convolutional neural network on programmable two-dimensional image processor

描述了一种方法，其包括在具有执行通道阵列和二维移位寄存器的图像处理器上执行卷积神经网络层。二维移位寄存器为执行通道提供局部相应的寄存器空间。卷积神经网络的执行包括将三维图像数据块的图像数据的平面加载到二维移位寄存器。执行卷积神经网络还包括通过依次进行以下步骤来执行图像数据的平面与系数值阵列的二维卷积：在执行通道内同时相乘相应的像素和系数值以产生部分乘积的阵列；在执行通道内同时将部分乘积和保存在图像数据内不同模板的二维寄存器中的部分乘积的相应累积值求和；以及通过移位二维移位寄存器阵列内的内容来实现执行通道内的二维卷积的值的对齐。

Method of AI-Based Diagnostic Technology Automation For Application To Equipment

본 발명은 자동차 내 회전체로부터 진동 데이타, 소음 데이터 및 CAN 데이터 중 어느 한 개 이상의 데이터를 입력받는 단계; 입력된 데이터를 트리밍하는 데이터 입력 처리 단계; 트리밍된 데이터로부터 특징(features)을 추출하는 단계; 진동 데이타, 소음 데이터 및 CAN 데이터 중 어느 한 개 이상의 데이터에 대한 하이퍼파라미터를 설정값을 정하는 단계; 개별 모델로서 ML(machine learning) 및 DL(deep learning)을 모두 포함하도록 총합 N개의 모델을 생성하고, N개의 상기 개별 모델에 대해 앙상블 모델 구조를 생성하는 단계;를 포함하며, N개의 상기 개별 모델의 비용함수값이 최소가 되도록 상기 하이퍼파라미터에 의한 상기 파라미터 업데이트가 진행됨에 따라, 보상이 최적화되면서 상기 앙상블 모델 구조를 이루는 N개의 상기 개별 모델 구조가 변경된다

Object detection method and device based on CNN (convolutional neural network) by utilizing 1x1 convolution

本发明涉及用于满足KPI的硬件优化的基于CNN的客体检测器的参数的学习方法，提供学习装置使第一转换层(Transposing Layer)或池化层串联(Concatenating)按建议的像素以生成整合特征图(Integrated Feature Map)，使第二转换层或池化层按像素分离利用整合特征图生成的体积得到调整的特征图以使分类层生成客体类信息的方法，由于在本发明中由同一处理器执行卷积运算及FC运算，因此能够减小芯片(Chip)的尺寸以优化硬件，从而能够满足KPI(Key Performance Index，关键绩效指标)。因此，具有无需在半导体制作过程中增设线路、节省电能、半导体模(Die)内用于将FC模块改为设置其他模块的空间增大等优点。

Training method of living body detection model, and method, device and equipment for living body detection

本公开提供了一种活体检测模型的训练方法和活体检测方法、装置，涉及人工智能技术领域，具体涉及计算机视觉和深度学习技术领域，可应用于人脸识别等场景。具体实现方案为：将包括目标对象的多个样本图像输入活体检测模型的特征提取网络，得到各样本图像的图像特征；各样本图像具有指示目标对象为真实类别的实际概率的标签；将图像特征输入活体检测模型的分类网络，得到目标对象为真实类别的第一预测概率；基于图像特征与预定特征序列中每个特征之间的相似度，确定目标对象为非真实类别的第二预测概率，该每个特征为标签指示的实际概率为零的第一样本图像的图像特征；基于第一预测概率、第二预测概率和实际概率，对活体检测模型进行训练。

Configurable accelerator frame apparatus and the system for depth convolutional neural networks

本公开涉及可配置的加速器框架设备和用于深度卷积神经网络的系统。实施例涉及包括串流开关和多个卷积加速器的可配置的加速器框架设备。串流开关具有多个输入端口和多个输出端口。输入端口中的每一个在运行时可配置，以经由串流链路将数据单向传递到输出端口中的任何一个或多个。多个卷积加速器中的每一个在运行时可配置，以经由多个串流开关输出端口中的至少两个来单向接收输入数据，并且多个卷积加速器中的每一个在运行时进一步可配置，以经由串流开关的输入端口单向地传送输出数据。

Audio encoding method, apparatus, computer device and medium

本申请是关于一种音频编码方法、装置、计算机设备及介质，属于音视频技术领域。该方法包括：获取原始音频中各个音频帧对应的音频特征参数；将所述音频特征参数输入编码码率预测模型中，得到所述编码码率预测模型输出的音频编码码率，其中，不同音频特征参数对应不同音频编码码率；基于所述音频编码码率对所述音频帧进行语音编码，并基于各帧音频帧对应的编码结果生成目标音频数据。通过分析原始音频中各个音频帧对应的音频特征参数，以实现基于音频特征参数动态调控音频编码码率的目的，可以为各个音频帧确定较适合的音频编码码率，从而提高整个音频的编码质量。

Processing communication signals using machine learning networks

公开用于使用机器学习网络处理通信信号的方法、系统和设备，其包含编码在计算机存储媒体上的计算机程序。在一些实施方案中，针对数据信号生成导频和数据信息。使用正交频分复用(OFDM)系统的调制器生成所述数据信号。通过通信信道发射所述数据信号以获得经修改导频和数据信息。使用机器学习网络处理所述经修改导频和数据信息。从所述机器学习网络获得与通过所述通信信道发射的所述数据信号相对应的预测。将所述预测与一组真实数据进行比较，并基于对应的误差项对所述机器学习网络应用更新。

System for automatic recognition and monitoring of vessel using artificial intelligence image processing and method for providing the same

본 발명은 인공지능 영상 처리를 이용한 선박 자동 인식 및 모니터링 시스템 및 그 제공 방법에 관한 것으로서, 더욱 상세하게는 헬리카이트와 같은 비행체를 통해 획득한 영상으로부터 해상에 존재하는 물체를 분류 및 인공지능 영상 처리를 수행하여 찾고자 하는 대상체를 정의하고, 대상체와 배경을 분류하고, 선박들의 위치를 지속적으로 송출하는 AIS와 연계하여 대상체 추적 가능 데이터로 출력함으로써, 외부 환경 변화에 따른 영상 데이터의 품질을 높이고, 해상의 현재 상황 감시, 선박의 자동 인식 및 추적, 및 위험 상황 알림이 가능한 인공지능 영상 처리를 이용한 선박 자동 인식 및 모니터링 시스템 및 그 제공 방법을 제공한다.

Neural network construction method and device

本申请公开了人工智能领域的神经网络构建方法以及装置，用于准确高效地构建目标神经网络，构建出的目标神经网络输出的准确度高，还可以应用于不同的应用场景中，泛化能力强。该方法包括：获取起点网络，该起点网络包括多个串行子网络；基于预设的第一搜索空间对起点网络进行至少一次变形，得到串行网络，第一搜索空间包括对起点网络进行变形使用的参数的范围；若串行网络满足预设条件，则通过预设的数据集对串行网络进行训练，得到训练后的串行网络；若训练后的串行网络满足终止条件，则根据训练后的串行网络得到目标神经网络。

Appratus and method for detecting object

영상 내의 오브젝트를 검출하기 위해, 원본 영상 및 해상도를 낮춘 영상들에 오브젝트 검출 필터를 이용하여 제1 오브젝트를 검출하고, 원본 영상 및 해상도를 낮춘 영상들 중 어느 하나로 결정된 타겟 영상을 입력으로 하는 뉴럴 네트워크 기반의 오브젝트 검출 모델을 이용하여 제2 오브젝트를 검출한다.

Method and apparatus for video coding

본 개시내용의 양태들은 비디오 처리에서와 같은 신경망 처리를 위한 방법들 및 장치들을 제공한다. 일부 예들에서, 신경망 처리를 위한 장치는 처리 회로를 포함한다. 처리 회로는 합성곱 연산에 대한 입력이 구분적으로 일정한 제1 입력 채널을 포함한다고 결정한다. 그 다음, 처리 회로는 합성곱 연산을 위한 입력의 다른 채널들에 기초하여 제1 중간 출력 채널을 계산하고; 그 다음에, 제1 중간 출력 채널 및 제1 입력 채널의 조합(예컨대, 선형 조합)에 기초하여 합성곱 연산의 출력을 생성한다.

Data processing method and related equipment

本申请涉及人工智能领域，公开了一种数据处理方法，包括：获取待处理数据以及目标神经网络模型，目标神经网络模型包括第一transformer层，第一transformer层包括第一残差支路和第二残差支路，第一残差支路包括第一注意力头，第二残差支路包括目标前馈层FFN；根据目标神经网络模型对待处理数据进行目标任务相关的处理，以得到数据处理结果，其中目标神经网络模型用于将第一注意力头的输出与第一权重值进行目标运算，得到第一残差支路的输出，和/或目标神经网络模型用于将目标FFN的输出与第二权重值进行目标运算，得到第二残差支路的输出。本实施例针对于不同的任务，设置了用于控制残差支路的输出的权重值，降低了终端设备运行目标神经网络模型的计算资源需求。

AUTOMATIC VISUAL MEDIA TRANSMISSION ERROR ASSESSMENT

A method or system is disclosed to assess transmission errors in a visual media input. Domain knowledge is obtained from the visual media input by content analysis, codec analysis, distortion analysis, and human visual system modeling. The visual media input is divided into partitions, which are passed into deep neural networks (DNNs). The DNN outputs of all partitions are combined with the guidance of domain knowledge to produce an assessment of the transmission error. In one or more illustrative examples, transmission error assessment at a plurality of monitoring points in a visual media communication system is collected and assessed, followed by quality control processes and statistical performance assessment on the stability of the visual communication system.

DETECTING AN IN-FIELD EVENT

Examples are disclosed that relate to methods, computing devices, and systems for detecting an in-field event. One example provides a method comprising, during a training phase, receiving one or more training data streams. The training data stream(s) include an audio input comprising a semantic indicator. The audio input is processed to recognize the semantic indicator. A subset of data is selected and used to train a machine learning model to detect the in-field event, and the method further comprises outputting the trained machine learning model. During a run-time phase, the method comprises receiving one or more run-time input data streams. The trained machine learning model is used to detect a second instance of the in-field event in the one or more run-time input data streams. The method further comprises outputting an indication of the second instance of the in-field event.

METHOD OF TRAINING NEURAL NETWORK MODEL FOR CALCULATING LEARNING ABILITY AND METHOD OF CALCULATING LEARNING ABILITY OF USER

Provided are a method of training a neural network for calculating a learning ability and a method of calculating a user's learning ability. The method of training a neural network includes acquiring an assessment database including data, which includes question information answered by a user at a second time point earlier than a first time point, the user's answer information to the question information, and the user's score information in a second assessment system, acquired from the second assessment system different from a first assessment system, generating an answer sequence from the assessment database by matching the answer information with the score information to prepare a training set, preparing a neural network for calculating the user's score information in the second assessment system on the basis of the answer information in the second assessment system, and training the neural network with the training set.

Single-frame stripe analysis method for generating antagonistic neural network based on multiple scales

本发明公开了一种基于多尺度生成对抗神经网络的单帧条纹分析方法，其包括构建多尺度生成对抗神经网络模型；构建多尺度生成对抗神经网络模型的综合损失函数L；采集多尺度生成对抗神经网络的训练数据，并利用训练数据对多尺度生成对抗神经网络进行训练；将待测条纹图像输入至训练好的多尺度图像生成器，获得对应的正弦项、余弦项和调制度图，利用反正切函数计算相位。本发明中的神经网络经训练好后，计算过程不需要人为地设置复杂的计算参数，操作更为简便。由于神经网络的输入为单幅条纹图像，本发明为运动物体的条纹分析提供了高效、高精度的相位计算方法。

Method and system for non-invasive gene detection using Artificial Intelligence (AI) models

使用基于人工智能(AI)的计算系统，在植入前非侵入性地估计在胚胎图像中是否存在一系列非整倍体和镶嵌体。具有相似不良结果风险的非整倍体和镶嵌体被分组，用它们的组标记训练图像。使用相同的训练数据集为每个组训练单独的AI模型，然后，例如通过使用系综或蒸馏方法将各单独的模型组合起来，以开发可以识别广泛的非整倍体风险和镶嵌风险的模型。通过训练多个模型(包括二元模型、分级分层模型和多类别模型)生成针对一个组的AI模型。具体地，分级分层模型是通过将质量标签分配给图像来生成的。在每一层，训练集被划分成质量最好的图像和其它图像。该层的模型在质量最好的图像上进行训练，其它图像被传递到下一层并重复该过程(于是，剩余的图像被分成下一个质量最好的图像和其它图像)。然后，最终模型可用于在植入前从胚胎图像非侵入性地识别非整倍体和镶嵌体以及相关联的不良后果风险。

Hierarchical learning of weights for neural networks performing multiple analyses

提供用于执行医学成像分析的系统和方法。输入医学成像数据被接收用于执行多个医学成像分析其中特定的一个。在使用被训练用于执行所述多个医学成像分析的神经网络的情况下生成输出，该输出提供关于输入医学成像数据的特定医学成像分析的结果。通过在使用与多个医学成像分析其中的不同的一个相关联的一个或多个权重的情况下学习与特定医学成像分析相关联的一个或多个权重来训练该神经网络。输出用于执行该特定医学成像分析的所生成的输出。

System and method for diagnosing small bowel preparation scale

본 발명은 소장(small bowel) 정결도를 진단하는 시스템에 관한 것으로, 다수의 소장 영상 중에서 유사한 소장 영상들의 대표 영상을 선별하도록 분석하는 유사도 분석부, 다수의 소장 영상을 학습한 상태에서 정결도를 진단하고자 하는 일련의 다수의 소장 영상을 수신하였을 시 상기 대표 영상을 학습 결과물에 적용하여 소장 정결도를 예측함으로써 점수별로 소장 정결도를 분류하는 영상 분류부 및 상기 대표 영상의 소장 정결도에 관한 점수 및 상기 대표 영상과 유사한 소장 영상의 개수를 기반으로 상기 일련의 다수의 소장 영상에 관한 최종 소장 정결도를 산출하는 정결도 진단부를 포함할 수 있다. The present invention relates to a system for diagnosing small bowel cleanliness, a similarity analysis unit that analyzes to select representative images of similar small intestine images from among multiple small bowel images, and a cleanliness level after learning multiple small bowel images When a series of multiple small intestine images to be diagnosed are received, an image classification unit that classifies small intestine cleanliness by score by applying the representative image to the learning result to predict the degree of cleanliness of the small intestine and the score on the degree of cleanliness of the small intestine of the representative image and a cleanliness diagnosis unit that calculates a final degree of cleanliness of the small intestine with respect to the series of multiple small intestine images based on the number of small intestine images similar to the representative image.

Method for learning a supervised artificial intelligence intended to identify a predetermined object in the environment of an aircraft

La présente invention concerne un procédé d’apprentissage d’une intelligence artificielle destinée à identifier un objet prédéterminé (20,25) dans l’environnement d’un aéronef en vol. Ledit procédé comporte des étapes d’identification d’au moins un objet prédéterminé (20,25) sur des représentations représentant au moins un objet prédéterminé (20,25) et son environnement, d’établissement d’un jeu d’apprentissage et d’un jeu de validation, ledit jeu d’apprentissage et ledit jeu de validation comprenant plusieurs représentations parmi lesdites représentations représentant au moins un objet prédéterminé (20,25), d’apprentissage de ladite intelligence artificielle avec ledit jeu d’apprentissage et de validation de ladite intelligence artificielle avec ledit jeu de validation. Ladite intelligence artificielle peut ensuite être utilisée dans un procédé d’aide à l’atterrissage d’un aéronef (1) pour l’identification d’une hélisurface (20,25) où effectuer ledit atterrissage. Ladite intelligence artificielle peut aussi être utilisée dans un procédé d’évitement d’un câble pour l’identification de câbles situés sur ou à proximité de la trajectoire dudit aéronef. Figure abrégé : figure 4 The present invention relates to a method of training an artificial intelligence intended to identify a predetermined object (20,25) in the environment of an aircraft in flight. Said method comprises steps of identifying at least one predetermined object (20,25) on representations representing at least one predetermined object (20,25) and its environment, of establishing a learning game and of a validation game, said training game and said validation game comprising several representations among said representations representing at least one predetermined object (20,25), for training said artificial intelligence with said training and validation game of said artificial intelligence with said validation set. Said artificial intelligence can then be used in a method for assisting the landing of an ...

Selecting patterns based on their representation characterized by lithographic apparatus or process characteristic

本文中描述用於選擇用於訓練或校準與半導體製造相關之模型之圖案的方法及設備。該方法包括：獲得一第一圖案集合；在一表示域中表示該第一圖案集合中之每一圖案，該表示域對應於電磁函數；及依據該表示域自該第一圖案集合選擇一第二圖案集合。

Improved VGG16 network pig identity recognition method based on transfer learning

本发明公开了一种基于迁移学习的改进VGG16网络猪的身份识别方法。先对处理好的视频进行逐帧提取，获得一系列图片，这些经过预处理成数据集，然后再进行划分训练集和测试集；构建改进的VGG16网络训练模型BN‑VGG16，保存预训练的特征提取模型Pre‑VGG16；接下来就是迁移学习过程，把源域训练获得的Pre‑VGG16特征提取网络迁移到用来识别猪的Pig‑Vgg16网络中；对调整尺寸后的数据集进行多分块改进的绝对值差分局部方向模式(Multi Block ImproveAbsolute Difference Local Direction Pattern,简称MB‑IADLDP)特征提取，并进行串行融合，最后进行猪的身份识别。基于迁移学习的改进VGG16模型在运行速度和精度上都优于传统的VGG16网络模型。