Empirical Evaluation of an Average Reward Learning Method Handling Simultaneous Learning Episodes in a Dynamic Environment with Emerging Tasks

ヴァルディヴィエルソ アレックス; 宮本 俊幸

doi:10.11509/sci.SCI10.0.259.0

The 54th Annual Conference of the Institute of Systems, Control and Information Engineers

Session ID : F234

DOI https://doi.org/10.11509/sci.SCI10.0.259.0

Conference information

Host: Institute of Systems, Control and Information Engineers

Empirical Evaluation of an Average Reward Learning Method Handling Simultaneous Learning Episodes in a Dynamic Environment with Emerging Tasks

*Alex Valdivielso, Toshiyuki Miyamoto

Author information

Keywords: Reinforcement Learning, Average Reward, Learning Episode, Dynamic Environment

CONFERENCE PROCEEDINGS FREE ACCESS

Details

Abstract

Average learning methods (ARLMs) show a poor performance in environments in which they must deal with several tasks simultaneously. In this paper we present the evaluation of an ARLM adapted to handle simultaneous learning episodes. We compare its performance against a conventional ARLM in a multicar elevator system.

Corresponding author

Register with J-STAGE for free!