Presentation 2002/1/22
Reinforcement Learning and Goal Estimation by Multiple Forward and Reward Models
Norikazu SUGIMOTO, Kazuyuki SAMEJIMA, Kenji DOYA, Mitsuo KAWATO,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) This reseach presents a new reinforcement learning framework, "Combinatorial Model-based Reinforcement Learning (CMRL)", which flexibly combines forward models, reward models, and contorollers. First, appropriate forward models and reward models are selected based on the correctness of their predictions. Then an appropriate controller is selected based on the TD-error given by the models and the controllers. A similar module selection method can be applied to imitation learning that takes into account the difference in the parameters of the learner and the teacher.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) module learning / reinforcement learning / nonlinear control / imitation learning
Paper #
Date of Issue

Conference Information
Committee NC
Conference Date 2002/1/22(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) Reinforcement Learning and Goal Estimation by Multiple Forward and Reward Models
Sub Title (in English)
Keyword(1) module learning
Keyword(2) reinforcement learning
Keyword(3) nonlinear control
Keyword(4) imitation learning
1st Author's Name Norikazu SUGIMOTO
1st Author's Affiliation NAra Institute of Science and Technology:ATR, Human Information Science Laboratories, Department 3:Creating the Brain, CREST, Japan Science and Technology Corporation()
2nd Author's Name Kazuyuki SAMEJIMA
2nd Author's Affiliation ATR, Human Information Science Laboratories, Department 3:Creating the Brain, CREST, Japan Science and Technology Corporation
3rd Author's Name Kenji DOYA
3rd Author's Affiliation NAra Institute of Science and Technology:ATR, Human Information Science Laboratories, Department 3:Creating the Brain, CREST, Japan Science and Technology Corporation
4th Author's Name Mitsuo KAWATO
4th Author's Affiliation NAra Institute of Science and Technology:ATR, Human Information Science Laboratories, Department 3
Date 2002/1/22
Paper #
Volume (vol) vol.101
Number (no) 616
Page pp.pp.-
#Pages 8
Date of Issue