서브메뉴
검색
Bounded Subgradient Trajectories in Semialgebraic Optimization
Bounded Subgradient Trajectories in Semialgebraic Optimization
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202103142
- ISBN
- 9798314886533
- DDC
- 519
- 저자명
- Li, Xiaopeng.
- 서명/저자
- Bounded Subgradient Trajectories in Semialgebraic Optimization
- 발행사항
- [Sl] : Columbia University, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 186 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-11, Section: B.
- 주기사항
- Advisor: Josz, Cedric.
- 학위논문주기
- Thesis (Ph.D.)--Columbia University, 2025.
- 초록/해제
- 요약Solving modern data science problems relies on optimization algorithms that often succeed in high-dimensional, large-scale settings. First-order methods, in particular, are widely used due to their low per-iteration cost and scalability. However, despite their empirical success, theoretical understanding of their behavior in such settings remains limited. Classical optimization theory is typically built on assumptions about the objective function, such as convexity, smoothness, and coercivity, which are rarely satisfied in practice. In addition, common assumptions on algorithms, such as the boundedness of iterates or the existence of limit points, are often difficult to verify. To address these challenges, we develop a framework that replaces these assumptions with easily checkable conditions, using tools from dynamical systems, semialgebraic geometry, and variational analysis.A key result of this thesis is that a broad class of optimization problems, including phase retrieval, matrix sensing, and neural networks, have bounded gradient flows. This property is central to multiple aspects of optimization. For landscape analysis, we develop practical tools for certifying the absence of bad local minima at infinity. For first-order algorithms, we analyze momentum methods and the proximal random reshuffling algorithm, proving global convergence of iterates and establishing improved convergence rates.
- 일반주제명
- Applied mathematics
- 일반주제명
- Engineering
- 키워드
- Data science
- 키워드
- Deep learning
- 키워드
- Dynamical system
- 키워드
- Machine learning
- 기타저자
- Columbia University Operations Research
- 기본자료저록
- Dissertations Abstracts International. 86-11B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017357168
■00520260202103142
■006m o d
■007cr#unu||||||||
■020 ▼a9798314886533
■035 ▼a(MiAaPQ)AAI31994503
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a519
■1001 ▼aLi, Xiaopeng.
■24510▼aBounded Subgradient Trajectories in Semialgebraic Optimization
■260 ▼a[Sl]▼bColumbia University▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a186 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-11, Section: B.
■500 ▼aAdvisor: Josz, Cedric.
■5021 ▼aThesis (Ph.D.)--Columbia University, 2025.
■520 ▼aSolving modern data science problems relies on optimization algorithms that often succeed in high-dimensional, large-scale settings. First-order methods, in particular, are widely used due to their low per-iteration cost and scalability. However, despite their empirical success, theoretical understanding of their behavior in such settings remains limited. Classical optimization theory is typically built on assumptions about the objective function, such as convexity, smoothness, and coercivity, which are rarely satisfied in practice. In addition, common assumptions on algorithms, such as the boundedness of iterates or the existence of limit points, are often difficult to verify. To address these challenges, we develop a framework that replaces these assumptions with easily checkable conditions, using tools from dynamical systems, semialgebraic geometry, and variational analysis.A key result of this thesis is that a broad class of optimization problems, including phase retrieval, matrix sensing, and neural networks, have bounded gradient flows. This property is central to multiple aspects of optimization. For landscape analysis, we develop practical tools for certifying the absence of bad local minima at infinity. For first-order algorithms, we analyze momentum methods and the proximal random reshuffling algorithm, proving global convergence of iterates and establishing improved convergence rates.
■590 ▼aSchool code: 0054.
■650 4▼aApplied mathematics
■650 4▼aEngineering
■653 ▼aData science
■653 ▼aDeep learning
■653 ▼aDynamical system
■653 ▼aMachine learning
■653 ▼aNonconvex optimization
■653 ▼aReal algebraic geometry
■690 ▼a0796
■690 ▼a0364
■690 ▼a0537
■71020▼aColumbia University▼bOperations Research.
■7730 ▼tDissertations Abstracts International▼g86-11B.
■790 ▼a0054
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17357168▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


