주 제:Generalized Hebbian Self-Organization and Restricted Boltzmann Machine
발표자: Prof. Shun-Ichi Amari (RIKEN Brain Science Institute, JAPAN)
일 시: 2013년 11월 6일(수) 15:00 ~ 16:00
장 소: 경북대학교 IT대학 4호관 101호
대 상: 경북대학교 교수, 대학원생 및 학부생
주최: BK+ Smart Life 실현을 위한 SW 인력양성사업단
강사약력:
- Current position
Senior Advisor, RIEKN Brain Science institute (RIKEN BSI), JAPAN
Senior Team Leader, Laboratory for Mathematical Neuroscience, RIKEN BSI
- Selected Professional Experience
Director, RIEKN Brain Science institute (RIKEN BSI), JAPAN (2003-2008)
Professor Emeritus, University of Tokyo, JAPAN (1996-present)
Professor, University of Tokyo, JAPAN (1981-1996)
- Selected Professional Affiliations and Honors
Chair of Kyoto Prize Committee (2004-2009)
IEEE Fellow (1994 – Present)
Professor Emeritus, University of Tokyo, JAPAN (1996-present)
President of IEICE (2004-2005)
- Selected Awards
Order of Cultural Merits 2013
Gabor Award (International Neural Networks Society) 2008
Special Award of Japanese Statistical Society 2002
IEEE Emanuel R. Piore Award 1997
Japan Academy Award 1995
IEEE Neural Networks Pioneer Award 1992
- More information: available at http://www.brain.riken.jp/en/faculty/details/2
내용요약:
Deep learning is a hot topic of research, since its performance has been proved outstanding. Lots of ingenious ideas are involved in it, so that it is difficult to understand why it works so well. We need theoretical elucidation why it works so well. Here we revisit a classical theory of self-organization of a layered network and its characteristics. It can be generalized to a self-organizing neural field. We study dynamics of self-organization of a neural field. We then compare it with the restricted Boltzmann machine, by studying dynamics of learning of a simple restricted Boltzmann machine. We further touch upon the comparison with the autoencoder. This talk does not give a completed theory but presents half-baked ideas on deep learning.
관련문의(초청자): 경북대학교 컴퓨터학부 박혜영 교수 (hypark@knu.ac.kr)