Presentation 2000/2/3
Reinforcement leaning capabilities of neural networks using the fluctuation-driven learning rule.
Noboru KUDO, Kuniyuki OKUDA, Takahumi OOHORI, Kazuhisa WATANABE,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) The TD(λ) method is a powerful strategy which enables a pair of neural networks(NNs), "actor" and "critic", to acquire the state control rules of dynamical environments on the basis of reinforcement signals. Through simulations of three tasks for swing up pole of Cart-Pole system and for keeping the double pendulum straight and upright, for raising the tip of double pendulum, we verified that a fluctuation-driven learning (FDL) rule made the multi-layered neural networks (MLNNs) applicable to the TD(λ) strategy. Furthermore, we discovered that the FDL rule enabled only one MLNN for actor to learn above tasks without TD(λ) strategy.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) fluctuation-driven learning / reinforcement learning / actor-critic / TD(λ) strategy simulation
Paper # NC99-83
Date of Issue

Conference Information
Committee NC
Conference Date 2000/2/3(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 leaning capabilities of neural networks using the fluctuation-driven learning rule.
Sub Title (in English)
Keyword(1) fluctuation-driven learning
Keyword(2) reinforcement learning
Keyword(3) actor-critic
Keyword(4) TD(λ) strategy simulation
1st Author's Name Noboru KUDO
1st Author's Affiliation Department of Electrical Engineering, Hokkaido Institute of Technology()
2nd Author's Name Kuniyuki OKUDA
2nd Author's Affiliation Department of Electrical Engineering, Hokkaido Institute of Technology
3rd Author's Name Takahumi OOHORI
3rd Author's Affiliation Department of Electrical Engineering, Hokkaido Institute of Technology
4th Author's Name Kazuhisa WATANABE
4th Author's Affiliation Department of Electrical Engineering, Hokkaido Institute of Technology
Date 2000/2/3
Paper # NC99-83
Volume (vol) vol.99
Number (no) 612
Page pp.pp.-
#Pages 8
Date of Issue