서브메뉴
검색
Multi-Selection in Voting and Recommendation: Approximation Algorithms and Convex Programming
Multi-Selection in Voting and Recommendation: Approximation Algorithms and Convex Programming
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202104733
- ISBN
- 9798290648934
- DDC
- 001
- 저자명
- Jiang, Zhihao.
- 서명/저자
- Multi-Selection in Voting and Recommendation: Approximation Algorithms and Convex Programming
- 발행사항
- [Sl] : Stanford University, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 133 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-03, Section: B.
- 주기사항
- Advisor: Goel, Ashish.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2025.
- 초록/해제
- 요약This thesis investigates algorithmic frameworks for multi-selection-selecting a small subset of alternatives to serve a population-across applications in multi-winner voting and differential privacy. The first two chapters focus on electing representative or stable committees in approval-based multi-winner voting, while the third leverages multi-selection to improve accuracy under input perturbation in a privacy-aware environment. All three contributions build on convex programming techniques to find principled, tractable solutions to these selection problems.In the first part, we study approximation algorithms for optimizing cost and welfare objectives under justified representation (JR) and its strengthened variants EJR+ and BJR. Prior work established that optimizing over JR committees is NP-hard, even for simple objectives. We design linear programming relaxations that capture JR constraints via carefully constructed variables, and develop randomized rounding techniques to obtain near-optimal committees. Specifically, we show that by relaxing JR slightly, allowing an approximate form of proportionality, we can achieve an O(log n) approximation to the minimum cost. Our methods extend naturally to stronger notions such as EJR+ and BJR, where we design specialized LP formulations that maintain proportionality guarantees under fractional voter-to-candidate assignments. These results match hardness lower bounds under standard complexity assumptions. Furthermore, we establish that under natural relaxations of committee size, simple greedy algorithms achieve nearly optimal welfare, and prove tight hardness of approximation results based on reductions from hypergraph problems. This chapter thus provides both algorithmic techniques and structural insights into the trade-off between fairness and quality in multi-selection problems.In the second part, we extend the notion of fairness to coalition stability by studying core outcomes in approval-based multi-selection. We introduce the concept of the Nash core, where voters are matched to candidates with proportional payments, and committees are selected to maximize a candidate-weighted Nash social welfare objective. We first show that in the fractional setting, there always exists a Nash-core solution, and we further design a proportional payment rule that assigns prices to candidates based on their approval ballots, ensuring that voters fairly contribute to committee members they support. Preliminary results for the discrete setting are also presented, where we define the discrete Nash core and propose heuristic algorithms for finding core-stable committees in practice. Empirical experiments on real voting data confirm that the Nash core approach leads to stable outcomes. By combining proportionality, coalition stability, and computational tractability, this chapter offers new directions for fair group decision making beyond traditional voting rules.In the third part, we incorporate privacy considerations into multi-selection by introducing a new system architecture for differentially private selection with multiple responses. Under this framework, users first perturb their private query values using local differential privacy mechanisms, and servers respond by sending back a small set of candidate results. We rigorously characterize the optimal mechanism under this architecture when the user feature space is one-dimensional and the utility loss is distance-based. Remarkably, we prove that adding Laplace noise to the user query, combined with a carefully optimized selection of k results from the server, minimizes expected disutility under privacy constraints. Our proof techniques involve developing a novel duality theory for differential-integral linear programs, which allows us to analyze infinite-dimensional optimization problems with both derivative and integral constraints. This duality framework may be of independent interest beyond privacy applications. Additionally, we show that the user's error decreases inversely with the number of returned results, achieving an O(1/k) trade-off between privacy and accuracy. This chapter blends ideas from differential privacy and optimization to design systems that respect both individual privacy and social utility.
- 일반주제명
- Software
- 일반주제명
- Linear programming
- 일반주제명
- Privacy
- 일반주제명
- Voters
- 일반주제명
- Public good
- 일반주제명
- Political advertising
- 일반주제명
- Elections
- 키워드
- Voting
- 기타저자
- Stanford University.
- 기본자료저록
- Dissertations Abstracts International. 87-03B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017358662
■00520260202104733
■006m o d
■007cr#unu||||||||
■020 ▼a9798290648934
■035 ▼a(MiAaPQ)AAI32149625
■035 ▼a(MiAaPQ)Stanfordbq150gy8422
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a001
■1001 ▼aJiang, Zhihao.
■24510▼aMulti-Selection in Voting and Recommendation: Approximation Algorithms and Convex Programming
■260 ▼a[Sl]▼bStanford University▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a133 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-03, Section: B.
■500 ▼aAdvisor: Goel, Ashish.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2025.
■520 ▼aThis thesis investigates algorithmic frameworks for multi-selection-selecting a small subset of alternatives to serve a population-across applications in multi-winner voting and differential privacy. The first two chapters focus on electing representative or stable committees in approval-based multi-winner voting, while the third leverages multi-selection to improve accuracy under input perturbation in a privacy-aware environment. All three contributions build on convex programming techniques to find principled, tractable solutions to these selection problems.In the first part, we study approximation algorithms for optimizing cost and welfare objectives under justified representation (JR) and its strengthened variants EJR+ and BJR. Prior work established that optimizing over JR committees is NP-hard, even for simple objectives. We design linear programming relaxations that capture JR constraints via carefully constructed variables, and develop randomized rounding techniques to obtain near-optimal committees. Specifically, we show that by relaxing JR slightly, allowing an approximate form of proportionality, we can achieve an O(log n) approximation to the minimum cost. Our methods extend naturally to stronger notions such as EJR+ and BJR, where we design specialized LP formulations that maintain proportionality guarantees under fractional voter-to-candidate assignments. These results match hardness lower bounds under standard complexity assumptions. Furthermore, we establish that under natural relaxations of committee size, simple greedy algorithms achieve nearly optimal welfare, and prove tight hardness of approximation results based on reductions from hypergraph problems. This chapter thus provides both algorithmic techniques and structural insights into the trade-off between fairness and quality in multi-selection problems.In the second part, we extend the notion of fairness to coalition stability by studying core outcomes in approval-based multi-selection. We introduce the concept of the Nash core, where voters are matched to candidates with proportional payments, and committees are selected to maximize a candidate-weighted Nash social welfare objective. We first show that in the fractional setting, there always exists a Nash-core solution, and we further design a proportional payment rule that assigns prices to candidates based on their approval ballots, ensuring that voters fairly contribute to committee members they support. Preliminary results for the discrete setting are also presented, where we define the discrete Nash core and propose heuristic algorithms for finding core-stable committees in practice. Empirical experiments on real voting data confirm that the Nash core approach leads to stable outcomes. By combining proportionality, coalition stability, and computational tractability, this chapter offers new directions for fair group decision making beyond traditional voting rules.In the third part, we incorporate privacy considerations into multi-selection by introducing a new system architecture for differentially private selection with multiple responses. Under this framework, users first perturb their private query values using local differential privacy mechanisms, and servers respond by sending back a small set of candidate results. We rigorously characterize the optimal mechanism under this architecture when the user feature space is one-dimensional and the utility loss is distance-based. Remarkably, we prove that adding Laplace noise to the user query, combined with a carefully optimized selection of k results from the server, minimizes expected disutility under privacy constraints. Our proof techniques involve developing a novel duality theory for differential-integral linear programs, which allows us to analyze infinite-dimensional optimization problems with both derivative and integral constraints. This duality framework may be of independent interest beyond privacy applications. Additionally, we show that the user's error decreases inversely with the number of returned results, achieving an O(1/k) trade-off between privacy and accuracy. This chapter blends ideas from differential privacy and optimization to design systems that respect both individual privacy and social utility.
■590 ▼aSchool code: 0212.
■650 4▼aSoftware
■650 4▼aLinear programming
■650 4▼aPrivacy
■650 4▼aVoters
■650 4▼aPublic good
■650 4▼aPolitical advertising
■650 4▼aElections
■653 ▼aJustified representation
■653 ▼aVoting
■653 ▼aConvex programming
■690 ▼a0796
■690 ▼a0454
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g87-03B.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17358662▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


