WO2024111736A1 - Procédé de traitement de données et dispositif informatique pour opération de convolution - Google Patents

Procédé de traitement de données et dispositif informatique pour opération de convolution Download PDF

Info

Publication number
WO2024111736A1
WO2024111736A1 PCT/KR2022/020226 KR2022020226W WO2024111736A1 WO 2024111736 A1 WO2024111736 A1 WO 2024111736A1 KR 2022020226 W KR2022020226 W KR 2022020226W WO 2024111736 A1 WO2024111736 A1 WO 2024111736A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
data format
integer
format
convolution operation
Prior art date
Application number
PCT/KR2022/020226
Other languages
English (en)
Korean (ko)
Inventor
이정승
Original Assignee
주식회사 모빌린트
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 주식회사 모빌린트 filed Critical 주식회사 모빌린트
Publication of WO2024111736A1 publication Critical patent/WO2024111736A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/15—Correlation function computation including computation of convolution operations
    • G06F17/153—Multidimensional correlation or convolution
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/15—Correlation function computation including computation of convolution operations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F7/00—Methods or arrangements for processing data by operating upon the order or content of the data handled
    • G06F7/38—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation
    • G06F7/48—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation using non-contact-making devices, e.g. tube, solid state device; using unspecified devices
    • G06F7/483—Computations with numbers represented by a non-linear combination of denominational numbers, e.g. rational numbers, logarithmic number system or floating-point numbers
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F7/00—Methods or arrangements for processing data by operating upon the order or content of the data handled
    • G06F7/38—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation
    • G06F7/48—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation using non-contact-making devices, e.g. tube, solid state device; using unspecified devices
    • G06F7/52—Multiplying; Dividing
    • G06F7/523—Multiplying only
    • G06F7/533—Reduction of the number of iteration steps or stages, e.g. using the Booth algorithm, log-sum, odd-even
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F7/00—Methods or arrangements for processing data by operating upon the order or content of the data handled
    • G06F7/38—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation
    • G06F7/48—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation using non-contact-making devices, e.g. tube, solid state device; using unspecified devices
    • G06F7/52—Multiplying; Dividing
    • G06F7/523—Multiplying only
    • G06F7/533—Reduction of the number of iteration steps or stages, e.g. using the Booth algorithm, log-sum, odd-even
    • G06F7/5334—Reduction of the number of iteration steps or stages, e.g. using the Booth algorithm, log-sum, odd-even by using multiple bit scanning, i.e. by decoding groups of successive multiplier bits in order to select an appropriate precalculated multiple of the multiplicand as a partial product
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F7/00—Methods or arrangements for processing data by operating upon the order or content of the data handled
    • G06F7/38—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation
    • G06F7/48—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation using non-contact-making devices, e.g. tube, solid state device; using unspecified devices
    • G06F7/544—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation using non-contact-making devices, e.g. tube, solid state device; using unspecified devices for evaluating functions by calculation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F7/00—Methods or arrangements for processing data by operating upon the order or content of the data handled
    • G06F7/38—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation
    • G06F7/48—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation using non-contact-making devices, e.g. tube, solid state device; using unspecified devices
    • G06F7/544—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation using non-contact-making devices, e.g. tube, solid state device; using unspecified devices for evaluating functions by calculation
    • G06F7/5443—Sum of products
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means

Definitions

  • the disclosed embodiments relate to data processing techniques for convolution operations.
  • a convolution operation model configured to perform a convolution operation on input data generally has the format of the input data fixed to a specific data format, but is obtained from a source device that provides data that is the target of the convolution operation. There are cases where the data does not match the input data format of the convolution computational model.
  • the input data format of the convolution operation model is a signed integer data format
  • the data acquired from the source device e.g., a camera
  • the process of converting the data obtained from the source device into real data format and then converting it back to the input data format of the convolution calculation model must be preceded.
  • the disclosed embodiments are intended to provide data processing technology for convolution operations.
  • a data processing method is performed on a computing device having one or more processors and a memory that stores one or more programs executed by the one or more processors, and includes a first integer data format. converting the data into second data having a second integer data format; Modifying preset weights and biases used for a convolution operation on data having the second integer data format, based on one or more preset parameters; and performing a convolution operation on the second data using the modified weight and bias.
  • the first integer data format may be an unsigned integer data format
  • the second integer data format may be a signed integer data format
  • the converting step may convert the first data into the second data by inverting the most significant bit (MSB) of the first data.
  • MSB most significant bit
  • the one or more preset parameters may include one or more parameters for converting data having a real number data format into data having the second integer data format.
  • the one or more preset parameters may include one or more normalization parameters for data normalization.
  • the one or more preset parameters may include one or more quantization parameters for quantizing data having the real data format and mapping it to data having the second integer data format.
  • a computing device includes one or more processors; and a memory storing one or more programs executed by the one or more processors, wherein the one or more processors convert first data having a first integer data format into second data having a second integer data format. and modifying preset weights and biases used for a convolution operation for data having the second integer data format, based on one or more preset parameters, and modifying the modified weights and biases.
  • a convolution operation is performed on the second data using .
  • the first integer data format may be an unsigned integer data format
  • the second integer data format may be a signed integer data format
  • the one or more processors may convert the first data into the second data by inverting the most significant bit (MSB) of the first data.
  • MSB most significant bit
  • the one or more preset parameters may include one or more parameters for converting data having a real number data format into data having the second integer data format.
  • the one or more preset parameters may include one or more normalization parameters for data normalization.
  • the one or more preset parameters may include one or more quantization parameters for quantizing data having the real data format and mapping it to data having the second integer data format.
  • the data format of the input data that is the target of the convolution operation does not match the data format for the convolution operation, it can be used for the convolution operation without a complex data format conversion process such as the prior art. It is possible to generate convolution operation results through simple modification of weights and bias. Accordingly, compared to the prior art, the time required for the convolution operation can be shortened and the efficiency of the convolution operation can be increased.
  • FIG. 1 is a configuration diagram of a data processing device according to an embodiment.
  • Figure 2 is a flowchart of a data processing method according to one embodiment.
  • FIG. 3 is a block diagram illustrating and illustrating a computing environment including a computing device according to an embodiment.
  • FIG. 1 is a configuration diagram of a data processing device according to an embodiment.
  • the data processing device 100 includes a data conversion unit 110, a weight conversion unit 120, and a calculation unit 130.
  • the data conversion unit 110, the weight conversion unit 120, and the calculation unit 130 are implemented using one or more physically separate devices, one or more processors, or a combination of one or more processors and software. It may be implemented, and unlike the example shown, specific operations may not be clearly distinguished.
  • the data conversion unit 110 converts first data having a first integer data format into second data having a second integer data format.
  • the first integer data format may be an unsigned integer ('uint') data format.
  • the first integer data type may be an unsigned integer data type expressed in 8 bits (i.e., 'uint8').
  • the size of the expression unit of the first integer data format is not limited to 8 bits, and may vary depending on the embodiment, for example, 16 bits, 32 bits, etc. (i.e., 'uint16', 'uint32', etc.) can do.
  • the first data may be, for example, image data acquired from a camera that outputs data in the 'uint' data format, but may also be acquired in real time or in advance from various devices that use the 'uint' data format. It may be acquired data.
  • the second integer data format may be a signed integer data format.
  • the second integer data type may be a signed integer data type represented by 8 bits (i.e., 'int8').
  • the size of the expression unit of the second integer data format is not limited to 8 bits, and may vary depending on the embodiment, for example, 16 bits, 32 bits, etc. (i.e., 'int16', 'int32', etc.) can do.
  • the size of the representation unit of the second integer data format may be the same as the size of the representation unit of the first integer data format.
  • the conversion unit 110 may convert the first data into second data in a second integer data format by inverting the most significant bit (MSB) of the first data.
  • MSB most significant bit
  • the conversion unit 110 can convert the first data into second data by inverting the MSB value of the first data (that is, converting 0 to 1 or converting 1 to 0).
  • the first data of the first integer data type with a value range of 0 to 2 n -1 is the second data of the second integer data type with a value range of -2 n-1 to 2 n-1 -1.
  • the conversion unit 110 converts the first data as shown in Table 1 below. 2 It can be converted into data.
  • the weight conversion unit 120 modifies preset weights and biases used for a convolution operation on data having a second integer data format, based on one or more preset parameters.
  • the preset weights and biases used for the convolution operation are the pre-trained weights and biases of an artificial neural network-based model configured to perform the convolution operation by receiving data having a second integer data format as input. It could be bias.
  • the artificial neural network-based model may be a convolutional neural network (CNN)-based model, but the structure of the artificial neural network-based model is not necessarily limited to a neural network with a specific structure.
  • the artificial neural network-based model may include, for example, weights and biases of a pre-trained model with values in a real data format (e.g., a floating point data format expressed in 32 bits). It may be a lightweight model created by converting from a real data format to a second integer data format through quantization.
  • a real data format e.g., a floating point data format expressed in 32 bits.
  • one or more preset parameters may include one or more parameters for converting data in a real number data format into data in a second integer data format.
  • the real number data format can be preset by the user.
  • the real data format may be a floating point data format expressed in 32 bits, but is not necessarily limited thereto and may be set in various ways depending on the embodiment.
  • one or more preset parameters may include one or more normalization parameters for data normalization.
  • data normalization may be for normalizing, for example, data in a preset real number data format, data in a first integer data format, etc. to a value within a preset range.
  • data subject to normalization is X
  • Equation 1 a and b are each preset normalization parameters and, depending on the embodiment, may be scalar values or vector values.
  • the one or more preset parameters may include one or more quantization parameters for quantizing data in a real data format and mapping it to data in a second integer data format.
  • Equation 2 floor() represents a function that finds the maximum integer value less than or equal to the real number in parentheses, and in Equation 3, round() represents a function that rounds the value below the decimal point and converts the real number in parentheses into an integer value.
  • ceil() represents a function that converts the real number in parentheses to an integer value by raising the value below the decimal point.
  • Q scale and Q zero represent quantization parameters.
  • Q scale represents a scale parameter and may mean the amount of change in a real number expressed in a real data format corresponding to the unit change (i.e., 1) of an integer value expressed in a second integer data format. .
  • Q zero represents a zero-point parameter and may mean an integer value corresponding to 0 expressed in a real number data format among integer values expressed in a second integer data format. That is, Q zero may be a value that expresses which integer value is mapped when 0 among the values expressed in the real number data format is converted to the second integer data format through quantization.
  • Q scale and Q zero can be determined based on the range of real values that can be expressed by the real number data format and the size (i.e., n) of the expression unit of the second integer data format, respectively.
  • Q scale can be determined according to Equations 5 and 6 below.
  • x max and x min represent the maximum and minimum values that can be expressed by real number data format, respectively.
  • the weight conversion unit 120 may modify the preset weights and biases for the convolution operation according to Equations 7 and 8 below, for example.
  • Weight new Weight/(a*Q scale )
  • Bias new Weight ⁇ (2 n-1 -b)/(a*Q scale )+Q zero ⁇ +Bias
  • Equations 7 and 8 'Weight' and 'Bias' represent the weight and bias before modification, respectively, and 'Weight new ' and 'Bias new ' represent the modified weight and bias, respectively.
  • the calculation unit 130 performs a convolution operation on the second data using the weight and bias modified by the weight conversion unit 120.
  • the convolution operation can be performed, for example, using Equation 9 below.
  • Equation 8 'Input' represents the second data, and 'Output' represents the result of the convolution operation.
  • the convolution operation may be performed using an artificial neural network-based model.
  • the calculation unit 130 converts the weights and biases of the artificial neural network-based model into the weights modified by the weight conversion unit 120. and bias can be changed. Thereafter, the calculation unit 130 may perform a convolution operation on the second data by using the second data as an input to an artificial neural network-based model with changed weights and biases.
  • Figure 2 is a flowchart of a data processing method according to one embodiment.
  • the method shown in FIG. 2 may be performed, for example, by the data processing device 100 described above.
  • the data processing device 100 converts first data having a first integer data format into second data having a second integer data format (210).
  • the first integer data format may be an unsigned integer data format
  • the second integer data format may be a signed integer data format
  • the data processing device 100 may convert the first data into a second integer data format by inverting the MSB of the first data.
  • the data processing device 100 modifies preset weights and biases used for a convolution operation on data having a second integer data format based on one or more preset parameters (220).
  • the one or more preset parameters may include one or more parameters for converting data in a real number data format into data in a second integer data format.
  • one or more preset parameters may include one or more normalization parameters for data normalization.
  • the one or more preset parameters may include one or more quantization parameters for quantizing data in a real data format and mapping it to data in a second integer data format.
  • the data processing device 100 performs a convolution operation on the second data using the modified weight and bias (230).
  • FIG. 3 is a block diagram illustrating and illustrating a computing environment including a computing device according to an embodiment.
  • each component may have different functions and capabilities in addition to those described below, and may include additional components in addition to those described below.
  • the illustrated computing environment 10 includes a computing device 12 .
  • Computing device 12 may be one or more components included in data processing device 100 according to one embodiment.
  • Computing device 12 includes one or more processors 14, a computer-readable storage medium 16, and a communication bus 18.
  • Processor 14 may cause computing device 12 to operate in accordance with the example embodiments noted above.
  • processor 14 may execute one or more programs stored on computer-readable storage medium 16.
  • the one or more programs may include one or more computer-executable instructions, which, when executed by the processor 14, cause computing device 12 to perform operations according to example embodiments. It can be.
  • the one or more processors 14 include at least one of a central processing unit (CPU), a graphics processing unit (GPU), and a neural processing unit. It may be included, but is not necessarily limited thereto.
  • Computer-readable storage medium 16 is configured to store computer-executable instructions or program code, program data, and/or other suitable form of information.
  • the program 20 stored in the computer-readable storage medium 16 includes a set of instructions executable by the processor 14.
  • computer-readable storage medium 16 includes memory (volatile memory, such as random access memory, non-volatile memory, or an appropriate combination thereof), one or more magnetic disk storage devices, optical disk storage devices, flash It may be memory devices, another form of storage medium that can be accessed by computing device 12 and store desired information, or a suitable combination thereof.
  • Communication bus 18 interconnects various other components of computing device 12, including processor 14 and computer-readable storage medium 16.
  • Computing device 12 may also include one or more input/output interfaces 22 and one or more network communication interfaces 26 that provide an interface for one or more input/output devices 24.
  • the input/output interface 22 and the network communication interface 26 are connected to the communication bus 18.
  • Input/output device 24 may be coupled to other components of computing device 12 through input/output interface 22.
  • Exemplary input/output devices 24 include, but are not limited to, a pointing device (such as a mouse or trackpad), a keyboard, a touch input device (such as a touchpad or touch screen), a voice or sound input device, various types of sensor devices, and/or imaging devices. It may include input devices and/or output devices such as display devices, printers, speakers, and/or network cards.
  • the exemplary input/output device 24 may be included within the computing device 12 as a component constituting the computing device 12, or may be connected to the computing device 12 as a separate device distinct from the computing device 12. It may be possible.
  • embodiments of the present invention may include a program for performing the methods described in this specification on a computer, and a computer-readable recording medium containing the program.
  • the computer-readable recording medium may include program instructions, local data files, local data structures, etc., singly or in combination.
  • the media may be specially designed and constructed for the present invention, or may be commonly used in the computer software field.
  • Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs, DVDs, and media specifically configured to store and perform program instructions such as ROM, RAM, flash memory, etc. Includes hardware devices.
  • Examples of the program may include not only machine language code such as that generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Analysis (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Computational Mathematics (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Nonlinear Science (AREA)
  • Neurology (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Image Processing (AREA)

Abstract

L'invention divulgue un procédé de traitement de données et un dispositif informatique pour une opération de convolution. Un procédé de traitement de données selon un mode de réalisation comprend les étapes consistant à : convertir des premières données ayant un premier format de données de type entier en secondes données ayant un second format de données de type entier ; modifier un poids et un biais prédéfinis qui sont utilisés pour appliquer une opération de convolution aux données ayant le second format de données de type entier sur la base d'un ou de plusieurs paramètres prédéfinis ; et appliquer l'opération de convolution aux secondes données à l'aide du poids et du biais modifiés.
PCT/KR2022/020226 2022-11-24 2022-12-13 Procédé de traitement de données et dispositif informatique pour opération de convolution WO2024111736A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
KR10-2022-0159369 2022-11-24
KR1020220159369A KR20240077167A (ko) 2022-11-24 2022-11-24 합성곱 연산을 위한 데이터 처리 방법 및 컴퓨팅 장치

Publications (1)

Publication Number Publication Date
WO2024111736A1 true WO2024111736A1 (fr) 2024-05-30

Family

ID=91196279

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/KR2022/020226 WO2024111736A1 (fr) 2022-11-24 2022-12-13 Procédé de traitement de données et dispositif informatique pour opération de convolution

Country Status (2)

Country Link
KR (1) KR20240077167A (fr)
WO (1) WO2024111736A1 (fr)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160026912A1 (en) * 2014-07-22 2016-01-28 Intel Corporation Weight-shifting mechanism for convolutional neural networks
US20170323197A1 (en) * 2016-05-03 2017-11-09 Imagination Technologies Limited Convolutional Neural Network Hardware Configuration
KR20200076800A (ko) * 2018-12-19 2020-06-30 고려대학교 산학협력단 출력 특징 맵의 0에 대한 연산을 스킵할 수 있는 합성곱 신경망의 연산 장치 및 그 동작 방법
KR20210093952A (ko) * 2018-11-15 2021-07-28 차나안 브라이트 사이트 컴퍼니 리미티드 적응 양자화 방법 및 장치, 장비, 매체
KR20220031117A (ko) * 2019-07-15 2022-03-11 페이스북 테크놀로지스, 엘엘씨 효율적인 곱셈을 위한 대안적인 숫자 형식을 지원하는 시스템 및 방법

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160026912A1 (en) * 2014-07-22 2016-01-28 Intel Corporation Weight-shifting mechanism for convolutional neural networks
US20170323197A1 (en) * 2016-05-03 2017-11-09 Imagination Technologies Limited Convolutional Neural Network Hardware Configuration
KR20210093952A (ko) * 2018-11-15 2021-07-28 차나안 브라이트 사이트 컴퍼니 리미티드 적응 양자화 방법 및 장치, 장비, 매체
KR20200076800A (ko) * 2018-12-19 2020-06-30 고려대학교 산학협력단 출력 특징 맵의 0에 대한 연산을 스킵할 수 있는 합성곱 신경망의 연산 장치 및 그 동작 방법
KR20220031117A (ko) * 2019-07-15 2022-03-11 페이스북 테크놀로지스, 엘엘씨 효율적인 곱셈을 위한 대안적인 숫자 형식을 지원하는 시스템 및 방법

Also Published As

Publication number Publication date
KR20240077167A (ko) 2024-05-31

Similar Documents

Publication Publication Date Title
US20240104378A1 (en) Dynamic quantization of neural networks
WO2022131497A1 (fr) Appareil d'apprentissage et procédé de génération d'image, et appareil et procédé de génération d'image
WO2023029349A1 (fr) Procédé et appareil de quantification de modèle, dispositif, support de stockage, produit-programme informatique et programme informatique
WO2020159016A1 (fr) Procédé d'optimisation de paramètre de réseau neuronal approprié pour la mise en œuvre sur matériel, procédé de fonctionnement de réseau neuronal et appareil associé
WO2022146080A1 (fr) Algorithme et procédé de modification dynamique de la précision de quantification d'un réseau d'apprentissage profond
WO2020189844A1 (fr) Procédé de traitement de réseau neuronal artificiel, et dispositif électronique pour cela
CN111078230A (zh) 一种代码生成方法和装置
WO2024111736A1 (fr) Procédé de traitement de données et dispositif informatique pour opération de convolution
WO2014142422A1 (fr) Procédé permettant de traiter un dialogue d'après une expression d'instruction de traitement, et appareil associé
WO2022139325A1 (fr) Système informatique pour apprentissage adaptatif multi-domaine basé sur un réseau neuronal unique sans sur-apprentissage, et procédé associé
WO2022114368A1 (fr) Procédé et dispositif de complétion de connaissances par représentation vectorielle continue d'une relation neuro-symbolique
CN113269319A (zh) 深度学习模型的调优方法、编译方法及计算装置
WO2022260392A1 (fr) Procédé et système pour générer un modèle de réseau neuronal artificiel de traitement d'image fonctionnant dans un terminal
WO2023249402A1 (fr) Procédé, programme, et appareil de traitement de données médicales pour l'entraînement d'un modèle d'apprentissage profond
WO2023277448A1 (fr) Procédé et système d'entraînement de modèle de réseau neuronal artificiel pour traitement d'image
CN113139650A (zh) 深度学习模型的调优方法和计算装置
WO2023003432A1 (fr) Procédé et dispositif pour déterminer une plage de quantification basée sur un taux de saturation pour la quantification d'un réseau de neurones
WO2022145713A1 (fr) Procédé et système d'allègement de modèle de réseau neuronal artificiel, et support d'enregistrement lisible par ordinateur non transitoire
WO2011007970A1 (fr) Procédé et appareil pour traiter une image
WO2012133976A1 (fr) Procédé, serveur, appareil de terminal, et support d'enregistrement lisible par ordinateur pour l'intégration d'une pluralité d'automates finis non déterministes pour la recherche par l'utilisation simultanée d'une pluralité d'expressions régulières
WO2023128024A1 (fr) Procédé et système de quantification de réseau d'apprentissage profond
WO2023177025A1 (fr) Procédé et appareil pour calculer un réseau neuronal artificiel sur la base d'une quantification de paramètre à l'aide d'une hystérésis
WO2019216551A1 (fr) Dispositif de génération de phrase de dialogue, robot de dialogue comprenant ce dernier et procédé de génération d'une phrase de dialogue
WO2024128631A1 (fr) Procédé de génération d'informations d'indice de matrice, et procédé de traitement de matrice utilisant des informations d'indice de matrice
WO2020213757A1 (fr) Procédé de détermination de similarité de mots

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22966596

Country of ref document: EP

Kind code of ref document: A1