컨텐츠 시작
학술대회/행사
초록검색
| 제출번호(No.) | 0130 |
|---|---|
| 분류(Section) | Special Session |
| 분과(Session) | (SS-19) Optimization and Machine Learning (SS-19) |
| 발표시간(Time) | 19th-A-10:30 -- 11:00 |
| 영문제목 (Title(Eng.)) |
Optimization algorithm design by continuous-time analysis |
| 저자(Author(s)) |
Jaewook J. Suh1 Seoul National University1 |
| 초록본문(Abstract) | After the continuous-time analysis of Nesterov's momentum-based acceleration introduced by Su et al., rich amount of research has been conducted to analyze first-order accelerated methods by the continuous-time perspective. Through such analyses, intuitive understandings of acceleration have been strengthened, and furthermore, discrete algorithms derived from continuous models have been proposed. In this talk, we explore the process of designing optimization algorithms through continuous-time analysis, highlighting the benefits of this approach. Moreover, we discuss the challenge of discretization that needs to be surmounted. |
| 분류기호 (MSC number(s)) |
90C25 |
| 키워드(Keyword(s)) | Optimization, accelerated method, continuous-time analysis |
| 강연 형태 (Language of Session (Talk)) |
English |