서브메뉴
검색
New Perspectives on Dimensionality Reduction and Selective Inference
New Perspectives on Dimensionality Reduction and Selective Inference
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202104848
- ISBN
- 9798288816208
- DDC
- 780
- 저자명
- Sood, Anav.
- 서명/저자
- New Perspectives on Dimensionality Reduction and Selective Inference
- 발행사항
- [Sl] : Stanford University, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 206 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-02, Section: B.
- 주기사항
- Advisor: Hastie, Trevor.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2025.
- 초록/해제
- 요약This thesis considers topics in two sub-fields of statistics---interpretable dimensionality reduction (Chapter 1 and Chapter 2) and selective inference (Chapter 3, Chapter 4, and Chapter 5). Interpretable dimensionality reduction concerns the problem of reducing a large dataset into a smaller one, while still ensuring the smaller dataset is readily interpretable. We consider the fundamental dimensionality reduction task of selecting the subset of variables that is most representative of the entire dataset. In Chapter 1, we show that two popular methods for this task, Column Subset Selection (CSS) and Principal Variables, are equivalent. This insight enables us to derive new methods for performing scalable CSS when (1) we have summary statistics in place of unit-level data, or (2) some of the unit-level data is missing and/or censored. In Chapter 2, we present a generative view of both of these methods. Specifically, we show that CSS and Principal Variables can both be interpreted as performing maximum likelihood estimation within a certain semi-parametric model. Within this model, we establish suitable conditions under which the CSS estimate is consistent in high dimensions, specifically in the proportional asymptotic regime where the number of variables over the sample size converges to a constant. Our generative viewpoint also leads to novel methodology for selecting the subset size for CSS/Principal Variables in a principled way. Selective inference concerns the problem of drawing inferences after selection of a data-dependent inferential question. In Chapter 3, we provide a simple and unifying framework for conditional selective inference that is centered on p-values. Specifically, we introduce selectively dominant p-values, a class of p-values that enable us to easily perform inference after selection. We give an exact characterization of when p-values are selectively dominant and argue that essentially all commonly used p-values are selectively dominant. In Chapter 4, we use our framework to study the inference on winners problem. We offer greater clarity on the behaviors and properties of existing inference on winners methods, while also generalizing them to new settings. We also identify a class of problems for which the conditional selective inference approach attains optimal power, but existing simultaneous inference approaches are powerless. In Chapter 5, we motivate a very broad and general class of rank verification problems that arise in a variety of different domains. We use our framework to derive a new inferential method for these problems that dominates existing simultaneous inference approaches, and we demonstrate our method's efficacy and versatility by applying it to a wide range of real-world problems.
- 일반주제명
- Music
- 일반주제명
- Musicians & conductors
- 일반주제명
- Personal profiles
- 일반주제명
- Pandemics
- 일반주제명
- Jazz
- 일반주제명
- Statistics
- 기타저자
- Stanford University.
- 기본자료저록
- Dissertations Abstracts International. 87-02B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017359198
■00520260202104848
■006m o d
■007cr#unu||||||||
■020 ▼a9798288816208
■035 ▼a(MiAaPQ)AAI32200931
■035 ▼a(MiAaPQ)Stanfordcf629qw1459
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a780
■1001 ▼aSood, Anav.
■24510▼aNew Perspectives on Dimensionality Reduction and Selective Inference
■260 ▼a[Sl]▼bStanford University▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a206 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-02, Section: B.
■500 ▼aAdvisor: Hastie, Trevor.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2025.
■520 ▼aThis thesis considers topics in two sub-fields of statistics---interpretable dimensionality reduction (Chapter 1 and Chapter 2) and selective inference (Chapter 3, Chapter 4, and Chapter 5). Interpretable dimensionality reduction concerns the problem of reducing a large dataset into a smaller one, while still ensuring the smaller dataset is readily interpretable. We consider the fundamental dimensionality reduction task of selecting the subset of variables that is most representative of the entire dataset. In Chapter 1, we show that two popular methods for this task, Column Subset Selection (CSS) and Principal Variables, are equivalent. This insight enables us to derive new methods for performing scalable CSS when (1) we have summary statistics in place of unit-level data, or (2) some of the unit-level data is missing and/or censored. In Chapter 2, we present a generative view of both of these methods. Specifically, we show that CSS and Principal Variables can both be interpreted as performing maximum likelihood estimation within a certain semi-parametric model. Within this model, we establish suitable conditions under which the CSS estimate is consistent in high dimensions, specifically in the proportional asymptotic regime where the number of variables over the sample size converges to a constant. Our generative viewpoint also leads to novel methodology for selecting the subset size for CSS/Principal Variables in a principled way. Selective inference concerns the problem of drawing inferences after selection of a data-dependent inferential question. In Chapter 3, we provide a simple and unifying framework for conditional selective inference that is centered on p-values. Specifically, we introduce selectively dominant p-values, a class of p-values that enable us to easily perform inference after selection. We give an exact characterization of when p-values are selectively dominant and argue that essentially all commonly used p-values are selectively dominant. In Chapter 4, we use our framework to study the inference on winners problem. We offer greater clarity on the behaviors and properties of existing inference on winners methods, while also generalizing them to new settings. We also identify a class of problems for which the conditional selective inference approach attains optimal power, but existing simultaneous inference approaches are powerless. In Chapter 5, we motivate a very broad and general class of rank verification problems that arise in a variety of different domains. We use our framework to derive a new inferential method for these problems that dominates existing simultaneous inference approaches, and we demonstrate our method's efficacy and versatility by applying it to a wide range of real-world problems.
■590 ▼aSchool code: 0212.
■650 4▼aMusic
■650 4▼aMusicians & conductors
■650 4▼aPersonal profiles
■650 4▼aPandemics
■650 4▼aJazz
■650 4▼aStatistics
■653 ▼aFundamental dimensionality reduction task
■653 ▼aColumn Subset Selection
■653 ▼aPrincipal Variables
■690 ▼a0413
■690 ▼a0463
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g87-02B.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17359198▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


