Summary
International Symposium on Nonlinear Theory and its Applications
2008
Session Number:B2L-C
Session:
Number:B2L-C4
A Profit Sharing Reinforcement Learning Method Using a Memory-Based Dynamic Reinforcement Function
Masaaki Usui, Hidehiro Nakano, Arata Miyauchi,
pp.-
Publication Date:2008/9/7
Online ISSN:2188-5079
Summary:
This paper proposes a new reinforcement learning method which decides dynamic reinforcement function. The conventional method has problems in deciding dynamic reinforcement function. In the proposed method, it is decided based on a memory of rules which an agent selects and executes. The proposed method provides better learning performances than the conventional SDPS. We present some numerical simulation results for static and dynamic maze environments.