Science Links Japan | MS-RL: Multi-Strategy Reinforcement Learning method for a learning agent under a variant environment.

MS-RL: Multi-Strategy Reinforcement Learning method for a learning agent under a variant environment.

Accession number;99A0263655

Title;MS-RL: Multi-Strategy Reinforcement Learning method for a learning agent under a variant environment.

Author; OKAMOTO MITSUYOSHI (Osaka Univ.) YAMAGUCHI TOMOHIRO (Osaka Univ.) YACHIDA MASAHIKO (Osaka Univ.)

Journal Title;IEIC Technical Report (Institute of Electronics, Information and Communication Engineers)

Journal Code:S0532B

ISSN:0913-5685

VOL.98;NO.499(AI98 68-75);PAGE.31-38(1999)

Figure&Table&Reference;FIG.11, REF.7

Pub. Country;Japan

Language;Japanese

Abstract;The object of this research is to realize a robust and flexible learning agent under a variant environment with intermittent changes of the learning conditions. Reinforcement learning is one of the possible behavior learning methods for an agent that behaves robustly in an unknown environment. Most previous reinforcement learning researches assume the limited conditions such as MDP environment to guarantee a rationality for learning, and tend to seek the convergence of the optimal learning result in infinite learning time. This paper presents Multi-Strategy Parallel Reinforcement Learning method(MSP-RL, in short) that performs the several different reinforcement learning algorithms in parallel. (author abst.)

Related Articles;