Processing

Please wait...

Settings

Settings

Goto Application

1. WO2020071618 - METHOD AND SYSTEM FOR ENTROPY-BASED NEURAL NETWORK PARTIAL LEARNING

Publication Number WO/2020/071618
Publication Date 09.04.2020
International Application No. PCT/KR2019/007639
International Filing Date 25.06.2019
IPC
G06N 3/04 2006.01
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 3/08 2006.01
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
08Learning methods
CPC
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 3/08
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
08Learning methods
Applicants
  • 인하대학교 산학협력단 INHA UNIVERSITY RESEARCH AND BUSINESS FOUNDATION [KR]/[KR]
Inventors
  • 강상길 KANG, Sang Gil
  • 허청환 HUR, Cheong Hwan
Agents
  • 이원희 LEE, Won Hee
Priority Data
10-2018-011773302.10.2018KR
Publication Language Korean (KO)
Filing Language Korean (KO)
Designated States
Title
(EN) METHOD AND SYSTEM FOR ENTROPY-BASED NEURAL NETWORK PARTIAL LEARNING
(FR) PROCÉDÉ ET SYSTÈME D'APPRENTISSAGE PARTIEL DE RÉSEAU NEURONAL BASÉ SUR L'ENTROPIE
(KO) 엔트로피 기반 신경망 부분학습 방법 및 시스템
Abstract
(EN)
The present invention relates to a method and system in which, for learning using convolutional neural networks, when a new class appears, a load of the learning can be reduced while maintaining accuracy. More specifically, the present invention relates to a method and system for partial learning of convolutional neural networks according to entropy-based weight evaluation, and the method of learning using neural networks comprises the steps of: (a) recognizing generation of a new class; (b) calculating a threshold value for determining qualitative information based on entropy of a plurality of weights and a weight for partial learning from among the plurality of weights; and (c) learning the new class by using weights, the qualitative information of which has a value equal to or smaller than the threshold value.
(FR)
La présente invention concerne un procédé et un système dans lesquels, pour un apprentissage à l'aide de réseaux neuronaux convolutionnels, lorsqu'une nouvelle classe apparaît, une charge de l'apprentissage peut être réduite tout en maintenant la précision. Plus spécifiquement, la présente invention concerne un procédé et un système d'apprentissage partiel de réseaux neuronaux convolutionnels selon une évaluation de poids basée sur l'entropie, et le procédé d'apprentissage à l'aide de réseaux neuronaux comprend les étapes consistant à : (a) reconnaître la génération d'une nouvelle classe ; (b) calculer une valeur de seuil pour déterminer des informations qualitatives sur la base d'une entropie d'une pluralité de poids et d'un poids pour un apprentissage partiel parmi la pluralité de poids ; et (c) apprendre la nouvelle classe en utilisant des poids dont les informations qualitatives ont une valeur égale ou inférieure à la valeur seuil.
(KO)
본 발명은 콘볼루션 신경망(Convolutional Neural Networks)을 이용한 학습에 있어서, 새로운 클래스가 나타난 경우 학습의 부하를 줄이면서도 정확도를 유지할 수 있는 방법 및 시스템에 관한 것으로, 더욱 상세하게는 엔트로피에 기반한 가중치 평가에 의한 콘볼루션 신경망의 부분학습 방법 및 시스템에 관한 것으로, 신경망(Neural Networks)을 이용한 학습 방법에 있어서, 새로운 클래스의 발생을 인식하는 (a) 단계; 복수의 가중치들의 엔트로피에 기반한 질적 정보 및 를 상기 복수의 가중치들 중 부분학습할 가중치를 결정하기 위한 임계값을 연산하는 (b) 단계; 및 상기 질적 정보가 임계값 이하의 값을 가지는 가중치들을 이용해 상기 새로운 클래스를 학습하는 (c) 단계;를 포함하는 구성을 개시한다.
Latest bibliographic data on file with the International Bureau