構造適応型Deep Belief Network事前学習を考慮した知識獲得の検討

URI http://harp.lib.hiroshima-u.ac.jp/pu-hiroshima/metadata/12597
File
Title
構造適応型Deep Belief Network事前学習を考慮した知識獲得の検討
Title Alternative
Knowledge Acquisition in Consideration of Pre-training for Adaptive Structural Learning of Deep Belief Network
Author
氏名 市村 匠
ヨミ イチムラ タクミ
別名 Ichimura Takumi
氏名 鎌田 真
ヨミ カマダ シン
別名 Kamada Shin
Abstract

Abstract—Deep Learning has a hierarchical network architecture to represent the complicated feature of in-put patterns. We have developed the adaptive structure learning method of Deep Belief Network (DBN) that can discover an optimal number of hidden neurons for given input data in a Restricted Boltzmann Machine (RBM)by neuron generation-annihilation algorithm, and hidden layers in DBN. The knowledge extraction method from the developed DBN and the rectification method of the signal flow on the wrong path have been developed. The fine-tuning method can reach an incredible high accuracy of classification (the best record). In Deep Learning, the layer-wise unsupervised pre-training can construct abstract and concrete modes of information processing. In this paper we improve the knowledge acquisition method to adopt a distinction between abstract and concrete. The empirical study was executed on the ChestX-ray8 database.

Description

開催日:平成30年7月28日
会場:広島工業大学

Journal Title
2018 IEEE SMC Hiroshima Chapter若手研究会講演論文集
Spage
70
Epage
76
Published Date
2018
Publisher
IEEE SMC Hiroshima Chapter
Contributor
市村匠
Language
jpn
NIIType
Conference Paper
Text Version
出版社版
Set
pu-hiroshima