주 제:Reinforcement Learning for Dynamic Environment in Maze

발표자: Prof. Tomoko Ozeki (Tokai University, Japan)

일 시: 2013년 11월 6일(수) 16:00 ~ 17:00

장 소: 경북대학교 IT대학 4호관 101호

대 상: 경북대학교 교수, 대학원생 및 학부생

주최: BK+ Smart Life 실현을 위한 SW 인력양성사업단

강사약력:

- Education

Tokyo Institute of Technology, Tokyo, Japan, 1986 - 1995

( Doctor of Science in Physics, 1995, Area of Statistical Physics and Neural Network

“Dynamics of Fully Connected Neural Network Model of Associative Memory”)

Master in Physics, 1992

B.A in Physics, 1990

- Career History

Professor at Tokai University (2011-Present)

Visiting Researcher at RIKEN Brain Science Institute (2005-Present)

Associate Professor at Tokai University, Japan (2005-2011)

Special Doctoral Researcher, BSI researcher at RIKEN BSI, Japan (1995-2005)

내용요약:

In this talk, I would like to introduce one of research activities in my laboratory. Reinforcement learning is an area of machine learning and is different from supervised learning where a teacher gives a correct answer. In reinforcement learning, an agent tries to find the series of optimal actions by trial and error in order to maximize the accumulated reward which is given by the environment. When we construct the reinforcement learning system, we do not have to specify the details of environment and the behavior of the agent. It is expected for the agent to adapt itself to the environment autonomously and it is possible in some extent. However, the reinforcement learning cannot deal with the sudden change of the environment. I introduce the concurrent Q-learning method proposed by Ollington and Vamplew, and our extension of their method.

관련문의(초청자): 경북대학교 컴퓨터학부 박혜영 교수 (hypark@knu.ac.kr)