서브메뉴
검색
Algorithms for Sparsity-Constrained Optimization Problems - On Polynomial Solvability and Approximability
Algorithms for Sparsity-Constrained Optimization Problems - On Polynomial Solvability and Approximability
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105120
- ISBN
- 9798291548028
- DDC
- 620
- 저자명
- Zhou, Dekun.
- 서명/저자
- Algorithms for Sparsity-Constrained Optimization Problems - On Polynomial Solvability and Approximability
- 발행사항
- [Sl] : The University of Wisconsin - Madison, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 161 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-02, Section: B.
- 주기사항
- Advisor: Del Pia, Alberto.
- 학위논문주기
- Thesis (Ph.D.)--The University of Wisconsin - Madison, 2025.
- 초록/해제
- 요약The optimization problem of finding a solution with a limited number of nonzero entries, known as a sparsity-constrained optimization problem, has garnered considerable attention; however, it is generally regarded as an NP-hard problem. This thesis investigates various types of sparsity-constrained optimization problems, such as the Sparse Integer Least Square problem (SILS) and the Sparse PCA problem (SPCA). Due to their computational complexity, it is not feasible to efficiently solve these problems with arbitrary inputs. Nevertheless, we present relaxation techniques that can be solved in polynomial time, providing provable approximation bounds to the original problems. Furthermore, we demonstrate that our proposed relaxations are sufficiently tight to solve the original problems for several input classes, and hence our results shed light on the applicability of these methods in real-world scenarios. In the first part of the thesis, we study SILS. Our study involves the derivation of an l1-based semidefinite programming (SDP) relaxation, along with a randomized algorithm that can handle a broader class of binary quadratic programs with cardinality constraint. We also establish sufficient conditions under which our SDP relaxation can successfully solve SILS, which we demonstrate through the analysis of two important special cases: the feature extraction problem and the integer sparse recovery problem. In the second part, we study SPCA. We introduce a randomized algorithm based on the basic SDP relaxation of SPCA, provide provable approximation bounds, and demonstrate that the basic SDP relaxation is resilient to adversarial perturbations in certain statistical models, and so do our algorithm. In the third part, we further our investigation into SPCA. We introduce a plug-and-play framework that accelerates most existing SPCA algorithms while incurring only a negligible loss in accuracy. The framework leverages a key structural property of SPCA: when the covariance matrix is block-diagonal, the global SPCA problem splits into independent SPCA sub-problems, allowing substantial speed-ups.
- 일반주제명
- Engineering
- 일반주제명
- Systems science
- 기타저자
- The University of Wisconsin - Madison Industrial Engineering
- 기본자료저록
- Dissertations Abstracts International. 87-02B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017359445
■00520260202105120
■006m o d
■007cr#unu||||||||
■020 ▼a9798291548028
■035 ▼a(MiAaPQ)AAI32238220
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a620
■1001 ▼aZhou, Dekun.
■24510▼aAlgorithms for Sparsity-Constrained Optimization Problems - On Polynomial Solvability and Approximability
■260 ▼a[Sl]▼bThe University of Wisconsin - Madison▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a161 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-02, Section: B.
■500 ▼aAdvisor: Del Pia, Alberto.
■5021 ▼aThesis (Ph.D.)--The University of Wisconsin - Madison, 2025.
■520 ▼aThe optimization problem of finding a solution with a limited number of nonzero entries, known as a sparsity-constrained optimization problem, has garnered considerable attention; however, it is generally regarded as an NP-hard problem. This thesis investigates various types of sparsity-constrained optimization problems, such as the Sparse Integer Least Square problem (SILS) and the Sparse PCA problem (SPCA). Due to their computational complexity, it is not feasible to efficiently solve these problems with arbitrary inputs. Nevertheless, we present relaxation techniques that can be solved in polynomial time, providing provable approximation bounds to the original problems. Furthermore, we demonstrate that our proposed relaxations are sufficiently tight to solve the original problems for several input classes, and hence our results shed light on the applicability of these methods in real-world scenarios. In the first part of the thesis, we study SILS. Our study involves the derivation of an l1-based semidefinite programming (SDP) relaxation, along with a randomized algorithm that can handle a broader class of binary quadratic programs with cardinality constraint. We also establish sufficient conditions under which our SDP relaxation can successfully solve SILS, which we demonstrate through the analysis of two important special cases: the feature extraction problem and the integer sparse recovery problem. In the second part, we study SPCA. We introduce a randomized algorithm based on the basic SDP relaxation of SPCA, provide provable approximation bounds, and demonstrate that the basic SDP relaxation is resilient to adversarial perturbations in certain statistical models, and so do our algorithm. In the third part, we further our investigation into SPCA. We introduce a plug-and-play framework that accelerates most existing SPCA algorithms while incurring only a negligible loss in accuracy. The framework leverages a key structural property of SPCA: when the covariance matrix is block-diagonal, the global SPCA problem splits into independent SPCA sub-problems, allowing substantial speed-ups.
■590 ▼aSchool code: 0262.
■650 4▼aEngineering
■650 4▼aSystems science
■653 ▼aApproximation algorithm
■653 ▼aMixed-integer optimization
■653 ▼aRandomized algorithm
■653 ▼aSparse optimization
■690 ▼a0796
■690 ▼a0537
■690 ▼a0790
■71020▼aThe University of Wisconsin - Madison▼bIndustrial Engineering.
■7730 ▼tDissertations Abstracts International▼g87-02B.
■790 ▼a0262
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17359445▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


