Processing

Please wait...

Settings

Settings

Goto Application

1. EP3557484 - NEURAL NETWORK CONVOLUTION OPERATION DEVICE AND METHOD

Office
European Patent Office
Application Number 17882134
Application Date 14.12.2017
Publication Number 3557484
Publication Date 23.10.2019
Publication Kind A4
IPC
G06N 3/04
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
04Architecture, e.g. interconnection topology
G06N 3/063
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
06Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
063using electronic means
CPC
G06N 3/0454
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
04Architectures, e.g. interconnection topology
0454using a combination of multiple neural nets
G06N 3/0635
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
06Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
063using electronic means
0635using analogue means
G06N 3/04
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
04Architectures, e.g. interconnection topology
G06N 3/063
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
06Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
063using electronic means
G06F 17/16
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
FELECTRIC DIGITAL DATA PROCESSING
17Digital computing or data processing equipment or methods, specially adapted for specific functions
10Complex mathematical operations
16Matrix or vector computation ; , e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
G06N 3/02
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
Applicants SHANGHAI CAMBRICON INF TECH CO LTD
Inventors CHEN YUNJI
ZHUANG YIMIN
LIU SHAOLI
GUO QI
CHEN TIANSHI
Designated States
Priority Data 201611152537 14.12.2016 CN
Title
(DE) VORRICHTUNG UND VERFAHREN FÜR FALTUNGSOPERATIONEN EINES NEURONALEN NETZES
(EN) NEURAL NETWORK CONVOLUTION OPERATION DEVICE AND METHOD
(FR) DISPOSITIF ET PROCÉDÉ POUR OPÉRATION DE CONVOLUTION DE RÉSEAU NEURONAL
Abstract
(EN)
The present disclosure provides a neural network convolution operation device and method for implementing a convolution operation of a weight matrix and neurons in a neural network by means of matrix multiplication. The method includes: shift operator, performing a winograd transformation on the neuron matrix and the weight matrix respectively to obtain a transformed neuron matrix and a transformed weight matrix; matrix multiplication operator, performing the matrix multiplication operation which multiplies the transformed neuron matrix and the transformed weight matrix together to obtain a multiplication matrix; the shift operator, performing a winograd inverse transformation on the multiplication matrix to obtain a convolution result; and controller, controlling the shift operator to perform matrix multiplication operation.

(FR)
La présente invention porte sur un dispositif et un procédé pour opération de convolution de réseau neuronal, le dispositif étant utilisé pour mettre en œuvre une opération de convolution d'une matrice de poids et de neurones dans un réseau neuronal au moyen d'un mode de multiplication matricielle, et comprenant : un opérateur de décalage, utilisé pour effectuer une transformation de Winograd sur une matrice de neurones et une matrice de poids respectivement de façon à obtenir une matrice de neurones transformée et une matrice de poids transformée ; un opérateur de multiplication matricielle, utilisé pour effectuer une opération de multiplication matricielle sur la matrice de neurones transformée et la matrice de poids transformée de façon à obtenir une matrice de multiplication, l'opérateur de décalage étant également utilisé pour effectuer une transformation de Winograd inverse sur la matrice de multiplication de façon à obtenir un résultat d'opération de convolution ; un dispositif de commande, utilisé pour commander l'opérateur de décalage pour effectuer la transformation de Winograd ou la transformation de Winograd inverse, et également utilisé pour commander l'opérateur de multiplication matricielle pour effectuer l'opération de multiplication matricielle.

Also published as
EP2017882134