Presentation | 2004/3/9 A Fast Learning Method for Reinforcement Learning on Shared Memory Multiprocessors Kouichirou MORI, Hayato YAMANA, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | In Reinforcement Learning, the agent learns by trial and error from a state without knowledge. Therefore, reinforcement learning has drawbacks that learning is slow. It is a serious problem how learns at high speed. In order to learn at high speed, some methods have been proposed. In the methods, the value function is divided. Then each divided value function is assigned to each processor, and updated in parallel. However, the method needs to exchange experiences frequently between the divided value function because of the character of reinforcement learning. It was a problem in the previous research that the overhead of communication between processors is large. In this paper, we propose the small overhead method that the parallel agents evaluate the shared value function asynchronously. As a result of the evaluation using shared memory type parallel machine, IBM pSeries 690, our method is approximately 22.2 times faster than sequential one in maze problem. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Reinforcement Learning / Speed-up Learning / Parallel Processing |
Paper # | AI2003-91 |
Date of Issue |
Conference Information | |
Committee | AI |
---|---|
Conference Date | 2004/3/9(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 | Artificial Intelligence and Knowledge-Based Processing (AI) |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | A Fast Learning Method for Reinforcement Learning on Shared Memory Multiprocessors |
Sub Title (in English) | |
Keyword(1) | Reinforcement Learning |
Keyword(2) | Speed-up Learning |
Keyword(3) | Parallel Processing |
1st Author's Name | Kouichirou MORI |
1st Author's Affiliation | Graduate School of Science and Engineering, Waseda University() |
2nd Author's Name | Hayato YAMANA |
2nd Author's Affiliation | Faculty of Science and Engineering, Waseda University |
Date | 2004/3/9 |
Paper # | AI2003-91 |
Volume (vol) | vol.103 |
Number (no) | 725 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |