Presentation 2012-12-12
Appropriate shaping rewards in reinforcement learning by aggregating transported states
Shinnosuke Oka, Kazushi Murakoshi,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Reinforcement learning adapting the idea of "Shaping" is a method to speed up the learning process by giving additional shaping reward that leads an agent from simple to complicated actions. Grzes and Kudenko(2010) proposed the online learning of shaping rewards by aggregating some states in the environment to one abstract state. Their method, however, used a position of the state in the state space to aggregate some states. This has a possibility that could aggregate some states which are far different from their values. In order to address that question, we propose a learning method which aggregates states the learning agent transported. We showed higher effectiveness of learning in a maze problem of our method than one in the conventional method.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) reinforcement learning / shaping reward / state aggregation
Paper # NC2012-74
Date of Issue

Conference Information
Committee NC
Conference Date 2012/12/5(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) Appropriate shaping rewards in reinforcement learning by aggregating transported states
Sub Title (in English)
Keyword(1) reinforcement learning
Keyword(2) shaping reward
Keyword(3) state aggregation
1st Author's Name Shinnosuke Oka
1st Author's Affiliation Toyohashi University of Technology()
2nd Author's Name Kazushi Murakoshi
2nd Author's Affiliation Toyohashi University of Technology
Date 2012-12-12
Paper # NC2012-74
Volume (vol) vol.112
Number (no) 345
Page pp.pp.-
#Pages 6
Date of Issue