서브메뉴
검색
Scalable Probabilistic Modeling and Inference With Structured Latent Representations
Scalable Probabilistic Modeling and Inference With Structured Latent Representations
Detailed Information
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202104736
- ISBN
- 9798290651675
- DDC
- 519.2
- 저자명
- Kruse, Liam A.
- 서명/저자
- Scalable Probabilistic Modeling and Inference With Structured Latent Representations
- 발행사항
- [Sl] : Stanford University, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 94 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-04, Section: B.
- 주기사항
- Advisor: Kochenderfer, Mykel.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2025.
- 초록/해제
- 요약The methods developed in this thesis advance the scalability of generative modeling and inference by using structured latent representations. High-dimensional distributions, such as those encountered in robotic trajectory prediction and image modeling, pose significant computational challenges for both learning and inference. By incorporating latent structure through low-rank approximations and change-of-variables techniques, we introduce principled ways to improve efficiency without sacrificing model expressivity. These contributions provide a foundation for scaling probabilistic modeling and inference to increasingly complex real-world domains.The first half of this thesis demonstrates how structured latent representations can enhance importance sampling for safety validation tasks. By conducting sampling in the latent space of a pre-trained normalizing flow, we mitigate issues of mode dropping and achieve better coverage of critical failure modes in high-dimensional systems. Additionally, we scale parametric importance sampling methods to higher dimensions by using low-rank mixture proposals. These techniques improve the reliability of black-box safety validation in applications ranging from autonomous racing to aircraft ground collision avoidance.The second half of this thesis explores structured latent representations beyond safety validation by improving normalizing flow training. By replacing simple latent priors with expressive low-rank mixture models, we reduce the burden on the flow to learn complex transformations from scratch. This approach warm-starts flow training, enhances model expressivity, and improves density estimation and generative sampling performance across tabular and image datasets.
- 일반주제명
- Probability
- 일반주제명
- Failure
- 일반주제명
- Normal distribution
- 일반주제명
- Applied mathematics
- 일반주제명
- Information science
- 기타저자
- Stanford University.
- 기본자료저록
- Dissertations Abstracts International. 87-04B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017358683
■00520260202104736
■006m o d
■007cr#unu||||||||
■020 ▼a9798290651675
■035 ▼a(MiAaPQ)AAI32149658
■035 ▼a(MiAaPQ)Stanfordgr246bp5721
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a519.2
■1001 ▼aKruse, Liam A.
■24510▼aScalable Probabilistic Modeling and Inference With Structured Latent Representations
■260 ▼a[Sl]▼bStanford University▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a94 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-04, Section: B.
■500 ▼aAdvisor: Kochenderfer, Mykel.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2025.
■520 ▼aThe methods developed in this thesis advance the scalability of generative modeling and inference by using structured latent representations. High-dimensional distributions, such as those encountered in robotic trajectory prediction and image modeling, pose significant computational challenges for both learning and inference. By incorporating latent structure through low-rank approximations and change-of-variables techniques, we introduce principled ways to improve efficiency without sacrificing model expressivity. These contributions provide a foundation for scaling probabilistic modeling and inference to increasingly complex real-world domains.The first half of this thesis demonstrates how structured latent representations can enhance importance sampling for safety validation tasks. By conducting sampling in the latent space of a pre-trained normalizing flow, we mitigate issues of mode dropping and achieve better coverage of critical failure modes in high-dimensional systems. Additionally, we scale parametric importance sampling methods to higher dimensions by using low-rank mixture proposals. These techniques improve the reliability of black-box safety validation in applications ranging from autonomous racing to aircraft ground collision avoidance.The second half of this thesis explores structured latent representations beyond safety validation by improving normalizing flow training. By replacing simple latent priors with expressive low-rank mixture models, we reduce the burden on the flow to learn complex transformations from scratch. This approach warm-starts flow training, enhances model expressivity, and improves density estimation and generative sampling performance across tabular and image datasets.
■590 ▼aSchool code: 0212.
■650 4▼aProbability
■650 4▼aFailure
■650 4▼aOrdinary differential equations
■650 4▼aNormal distribution
■650 4▼aApplied mathematics
■650 4▼aInformation science
■653 ▼aScaling probabilistic modeling
■653 ▼aStructured latent representations
■690 ▼a0364
■690 ▼a0723
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g87-04B.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17358683▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
Preview
Export
ChatGPT Discussion
AI Recommended Related Books
Подробнее информация.
- Бронирование
- не существует
- моя папка
- Первый запрос зрения
- Non-Book Loan Application
- Nighttime Book Loan Application
Available after logging in.


