서브메뉴
검색
Advancing Computations for Wasserstein Gradient Flows
Advancing Computations for Wasserstein Gradient Flows
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105305
- ISBN
- 9798265435071
- DDC
- 519
- 저자명
- Zuo, Xinzhe.
- 서명/저자
- Advancing Computations for Wasserstein Gradient Flows
- 발행사항
- [Sl] : University of California, Los Angeles, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 198 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-05, Section: B.
- 주기사항
- Advisor: Osher, Stanley J.
- 학위논문주기
- Thesis (Ph.D.)--University of California, Los Angeles, 2025.
- 초록/해제
- 요약Wasserstein gradient flows have emerged as a powerful mathematical framework that unifies optimization, sampling, and the dynamics of probability distributions. Rooted in optimal transport theory, they describe the steepest descent of energy functionals in the space of probability measures, providing deep geometric insight into algorithms used across machine learning, scientific computing, and statistical physics. As applications grow in scale and complexity, however, advancing the computational frontiers of Wasserstein gradient flows demands both new algorithmic formulations and rigorous numerical analysis. This dissertation aims to address these challenges by developing efficient, theoretically grounded methods that connect optimization dynamics, neural approximation, and stochastic sampling within the Wasserstein framework.We first propose Primal-Dual Damping (PDD) algorithms for optimization, which generalize the primal-dual hybrid gradient method and reveal a continuous-time limit governed by second-order dynamics that extend heavy-ball and Hessian-driven damping systems. Next, we analyze neural network-projected schemes for approximating one-dimensional Wasserstein gradient flows, establishing well-posedness, consistency, and error bounds for their neural representations and validating them on canonical PDEs such as Fokker-Planck and porous medium equations. Finally, we introduce Gradient-Adjusted Underdamped Langevin (GAUL) dynamics, which incorporate PDD mechanisms from optimization into stochastic sampling to achieve faster convergence toward target distributions. Together, these works contribute to both the theory and computation of Wasserstein gradient flows, bridging optimization, PDE, and sampling through a unified and rigorous mathematical perspective.
- 일반주제명
- Applied mathematics
- 일반주제명
- Mathematics
- 일반주제명
- Computer science
- 기타저자
- University of California, Los Angeles Mathematics 0540
- 기본자료저록
- Dissertations Abstracts International. 87-05B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017360109
■00520260202105305
■006m o d
■007cr#unu||||||||
■020 ▼a9798265435071
■035 ▼a(MiAaPQ)AAI32283108
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a519
■1001 ▼aZuo, Xinzhe.
■24510▼aAdvancing Computations for Wasserstein Gradient Flows
■260 ▼a[Sl]▼bUniversity of California, Los Angeles▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a198 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-05, Section: B.
■500 ▼aAdvisor: Osher, Stanley J.
■5021 ▼aThesis (Ph.D.)--University of California, Los Angeles, 2025.
■520 ▼aWasserstein gradient flows have emerged as a powerful mathematical framework that unifies optimization, sampling, and the dynamics of probability distributions. Rooted in optimal transport theory, they describe the steepest descent of energy functionals in the space of probability measures, providing deep geometric insight into algorithms used across machine learning, scientific computing, and statistical physics. As applications grow in scale and complexity, however, advancing the computational frontiers of Wasserstein gradient flows demands both new algorithmic formulations and rigorous numerical analysis. This dissertation aims to address these challenges by developing efficient, theoretically grounded methods that connect optimization dynamics, neural approximation, and stochastic sampling within the Wasserstein framework.We first propose Primal-Dual Damping (PDD) algorithms for optimization, which generalize the primal-dual hybrid gradient method and reveal a continuous-time limit governed by second-order dynamics that extend heavy-ball and Hessian-driven damping systems. Next, we analyze neural network-projected schemes for approximating one-dimensional Wasserstein gradient flows, establishing well-posedness, consistency, and error bounds for their neural representations and validating them on canonical PDEs such as Fokker-Planck and porous medium equations. Finally, we introduce Gradient-Adjusted Underdamped Langevin (GAUL) dynamics, which incorporate PDD mechanisms from optimization into stochastic sampling to achieve faster convergence toward target distributions. Together, these works contribute to both the theory and computation of Wasserstein gradient flows, bridging optimization, PDE, and sampling through a unified and rigorous mathematical perspective.
■590 ▼aSchool code: 0031.
■650 4▼aApplied mathematics
■650 4▼aMathematics
■650 4▼aComputer science
■653 ▼aOptimal transport theory
■653 ▼aPrimal-Dual Damping
■653 ▼aAlgorithmic formulations
■653 ▼aWasserstein gradient flows
■690 ▼a0364
■690 ▼a0984
■690 ▼a0405
■71020▼aUniversity of California, Los Angeles▼bMathematics 0540.
■7730 ▼tDissertations Abstracts International▼g87-05B.
■790 ▼a0031
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360109▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


