컨텐츠 시작
학술대회/행사
초록검색
제출번호(No.) | 0125 |
---|---|
분류(Section) | Special Session |
분과(Session) | (SS-16) Computational Methods for PDEs and Dynamical Systems (SS-16) |
발표시간(Time) | 20th-C-10:30 -- 11:00 |
영문제목 (Title(Eng.)) |
On the stability of Lipschitz continuous control problems and its application to reinforcement learning |
저자(Author(s)) |
Namkyeong Cho1, Yeoneung Kim2 POSTECH1, Seoul National University of Science and Technology2 |
초록본문(Abstract) | In this paper, we study the stability property of the Hamilton--Jacobi--Bellman (HJB) equation arising in reinforcement learning. It has been studied that Lipschitz continuous optimal control problems can be solved via a model-free off-line learning framework that leaves several questions on the stability properties on the HJB equation concerned. We focus on understanding , Moreover, we introduce a generalized form of HJB equation that can be implemented in the machine learning approach. It is also demonstrated by numerical examples. |
분류기호 (MSC number(s)) |
49L12, 68T07 |
키워드(Keyword(s)) | HJB equations, reinforcement learning |
강연 형태 (Language of Session (Talk)) |
Korean |