ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper 아타리 게임에 대한 내재적 보상을 이용한 강화학습 알고리즘
Cited - time in scopus Share share facebook twitter linkedin kakaostory
Authors
박성윤, 이상광
Issue Date
2022-06
Citation
대한전자공학회 학술 대회 (하계) 2022, pp.825-828
Publisher
대한전자공학회
Language
Korean
Type
Conference Paper
Abstract
As generating intrinsic reward in reinforcement learning has been advanced recently, one of the intrinsic rewards DoWhaM showed the greatest score in Corridor of MiniGrid. DoWhaM gives additional reward to a taken action which let agent go to a different state. However, environments the intrinsic reward was applied were mostly grid-based maze environment, MiniGrid. In this paper, we employ DoWhaM in Atari-2600 game and show the practical application of intrinsic reward. Also, we describe how the intrinsic reward affect to convergence in finding an optimal policy.
KSP Keywords
Grid-based, Intrinsic rewards, Optimal policy, Reinforcement Learning(RL), maze environment, practical application