Presentation 2001/6/22
Control of exploration and exploitation in reinforcement learning
Wako Yoshida, Shin Ishii,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Reinforcement learning (RL) is a learning framework based on trial-and-error, in which an agent learns by interacting with the environment. Model-based RL calculates the value function by modeling the environment; it is suitable for complicated environments. In this report, we propose a Bayes approximation method for the environmental model. One of the major issues in RL is the balance between exploration for searching for better control, and exploitation for obtaining the large reward. In order to control this contradictory balance, we introduce a control mechanism for the inverse temperature and an exploration bonus term, to the action selection. When our learning method is applied to a two-dimensional maze task, experimental results show taht the learning agent is able to adapt well to changes in the environment.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Model-based reinforcement learning / Bayes inference / exploration-exploitation problem / inverse temperature / exploration bonus
Paper # NC2001-28
Date of Issue

Conference Information
Committee NC
Conference Date 2001/6/22(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Neurocomputing (NC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Control of exploration and exploitation in reinforcement learning
Sub Title (in English)
Keyword(1) Model-based reinforcement learning
Keyword(2) Bayes inference
Keyword(3) exploration-exploitation problem
Keyword(4) inverse temperature
Keyword(5) exploration bonus
1st Author's Name Wako Yoshida
1st Author's Affiliation Nara Institute of Science and Technology:CREST Doya Project, Japan Science and Technology Corporation()
2nd Author's Name Shin Ishii
2nd Author's Affiliation Nara Institute of Science and Technology:CREST Doya Project, Japan Science and Technology Corporation
Date 2001/6/22
Paper # NC2001-28
Volume (vol) vol.101
Number (no) 154
Page pp.pp.-
#Pages 8
Date of Issue