지원사업
학술연구/단체지원/교육 등 연구자 활동을 지속하도록 DBpia가 지원하고 있어요.
커뮤니티
연구자들이 자신의 연구와 전문성을 널리 알리고, 새로운 협력의 기회를 만들 수 있는 네트워킹 공간이에요.
논문 기본 정보
- 자료유형
- 학위논문
- 저자정보
- 지도교수
- 이형옥
- 발행연도
- 2023
- 저작권
- 순천대학교 논문은 저작권에 의해 보호받습니다.
이용수7
이 논문의 연구 히스토리 (2)
- 이 논문의 후속연구가 궁금하신가요?
- 연관 학술논문 또는 학술발표를 통해 보다 발전된 연구결과를 확인하실 수 있습니다.
- 이 논문의 연구 히스토리 확인하기
초록· 키워드
오류제보하기
Various studies are being conducted as deep learning and reinforcement learning are likely to solve real-life problems while solving difficult problems in various fields such as go and games. This paper proposes a method of finding a winning strategy using the algorithm of reinforcement learning in Nim game, one of the combination games in which mathematical winning strategies exist.Using the representative Deep Q-network algorithm of value-based agents, the representative REINFORCE algorithm of policy-based agents, and the Actor-Critical algorithm combining the two, it was compared and analyzed through confrontation with Nim game must-win strategy agents. The victory strategy expressed the network value that selects the behavior in each algorithm as a graph and compared it with the victory strategy of the Nim game. The number of starts was divided into 10, 31, and 100, and 10 and 31 agents were well learned, but 100 agents were not learned.
As a result of the trial, all three agents were trained by randomly setting the starting number. The Actor-Critic agent was learned
the fastest, and the REINFORCE agent was learned the slowest. The competition between reinforcement learning agents was conducted with DQN agents versus DQN agents and DQN agents versus Actor-Critic agents, but there was no significant difference between DQN agents and Actor-Critic agents, and DQN agents found a winning strategy faster.
As a result of the trial, all three agents were trained by randomly setting the starting number. The Actor-Critic agent was learned
the fastest, and the REINFORCE agent was learned the slowest. The competition between reinforcement learning agents was conducted with DQN agents versus DQN agents and DQN agents versus Actor-Critic agents, but there was no significant difference between DQN agents and Actor-Critic agents, and DQN agents found a winning strategy faster.
목차
Ⅰ. 서 론 ········································································································· 11. 연구의 필요성 및 목적 ·············································································· 12. 연구 문제 ······································································································ 2Ⅱ. 이론적 배경 ·························································································· 41. 님게임의 필승전략 ······················································································ 41) 공정한 조합 게임 ··················································································· 42) 님게임의 정의 ························································································· 53) 님게임의 필승전략 ················································································· 52. 강화학습 ········································································································ 61) 강화학습의 정의와 종류 ······································································· 62) DQN(Deep Q-Network) 알고리즘 ···················································· 83) REINFORCE(Monte-Carlo policy gradient) 알고리즘 ··············· 94) Actor-Critic 알고리즘 ······································································· 11Ⅲ. 연구 방법 및 절차 ········································································ 121. 시뮬레이션 환경 및 구성 ······································································· 122. 강화학습 에이전트와 필승전략 에이전트 대결 ································· 123. 강화학습 에이전트끼리 대결 ································································· 14-i-- ii -Ⅳ. 연구 결과 분석 ················································································ 151. 강화학습 알고리즘 미적용 무작위 ( 선택 에이전트) ·························· 152. DQN 에이전트 ·························································································· 173. REINFORCE 에이전트 ············································································ 224. Actor-Critic 에이전트 ············································································ 255. DQN 에이전트와 DQN 에이전트 ························································· 266. DQN 에이전트와 Actor-Critic 에이전트 ··········································· 28Ⅴ. 결론 및 제언 ····················································································· 30참 고 문 헌 ······································································································· 32
최근 본 자료
댓글(0)
0