서브메뉴
검색
Stability and Efficiency in Multi-Agent Systems
Stability and Efficiency in Multi-Agent Systems
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105001
- ISBN
- 9798290651446
- DDC
- 153.8
- 서명/저자
- Stability and Efficiency in Multi-Agent Systems
- 발행사항
- [Sl] : Stanford University, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 190 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-04, Section: B.
- 주기사항
- Advisor: Saban, Daniela.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2025.
- 초록/해제
- 요약This thesis investigates the stability and efficiency of multi-agent systems, focusing on two main application areas: algorithmic competition in online marketplaces and the design of ranking mechanisms for preference aggregation. Although these domains arise in different settings, they are united by a common underlying challenge: the overall welfare and performance of the system crucially depend on the interactions among heterogeneous agents-each with their own preferences, and potentially strategic behavior.Analyzing the outcomes that emerge from the interactions of decentralized agents-and understanding their implications for overall welfare-is a central and challenging problem at the intersection of computer science, economics, and operations research. These challenges become especially pronounced in algorithmically powered environments, where systems are non-stationary, and agents have limited information, adapting their behavior in real-time. In such dynamic settings, traditional equilibrium and welfare analysis is often not applicable, opening the way to the definition and analysis of new equilibrium and welfare metrics. In the context of preference aggregation, traditional aggregation models have assumed that disagreements between users' preferences carry equal intensity regardless of the position, and that all alternatives are interchangeable. These assumptions may be analytically convenient, but they are increasingly implausible in modern applications such as recommender systems, where users' preferences are aggregated to provide new recommendations. For instance, in recommender systems, users pay far more attention to top-ranked suggestions than to lower ones-consider for example the difference between the first and tenth movie on a streaming platform-and alternatives often differ in intrinsic characteristics, such as genre, brand, or quality tier. Understanding how to measure differences in users' preferences accounting for these features becomes essential to devise effective rank aggregation mechanisms.The first chapter of my thesis is based on my work Equilibria with Dynamic Benchmarks in NonStationary Multi-Agent Systems,co-authored with Yonatan Gur and Bar Light. In this paper, we formulate and study a general time-varying multi-agent system where players repeatedly compete under incomplete information. Our work is motivated by scenarios commonly observed in online advertising and retail marketplaces, where agents and platform designers optimize algorithmic decision-making in dynamic competitive settings. In these systems, no-regret algorithms that provide guarantees relative to staticbenchmarks can perform poorly and the distributions of play that emerge from their interaction do not correspond anymore to static solution concepts such as coarse correlated equilibria. Instead, we analyze the interaction of dynamic benchmarkconsistent policies that have performance guarantees relative to dynamicsequences of actions, and through a novel tracking errornotion we delineate when their empirical joint distribution of play can approximate an evolving sequence of static equilibria. In systems that change sufficiently slowly (sub-linearly in the horizon length), we show that the resulting distributions of play approximate the sequence of coarse correlated equilibria, and apply this result to establish improved welfare bounds for smooth games. On a similar vein, we formulate internal dynamic benchmark consistent policies and establish that they approximate sequences of correlated equilibria. Our findings therefore suggest that, in a broad range of multi-agent systems where non-stationarity is prevalent, algorithms designed to compete with dynamic benchmarks can improve both individual and welfare guarantees, and their emerging dynamics approximate a sequence of static equilibrium outcomes.
- 일반주제명
- Motivation
- 일반주제명
- Integer programming
- 일반주제명
- Decision making
- 일반주제명
- Benchmarks
- 일반주제명
- Prices
- 일반주제명
- Computer 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■001000017359280
■00520260202105001
■006m o d
■007cr#unu||||||||
■020 ▼a9798290651446
■035 ▼a(MiAaPQ)AAI32149760
■035 ▼a(MiAaPQ)Stanfordzc980rv1768
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a153.8
■1001 ▼aCrippa, Ludovico.
■24510▼aStability and Efficiency in Multi-Agent Systems
■260 ▼a[Sl]▼bStanford University▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a190 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-04, Section: B.
■500 ▼aAdvisor: Saban, Daniela.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2025.
■520 ▼aThis thesis investigates the stability and efficiency of multi-agent systems, focusing on two main application areas: algorithmic competition in online marketplaces and the design of ranking mechanisms for preference aggregation. Although these domains arise in different settings, they are united by a common underlying challenge: the overall welfare and performance of the system crucially depend on the interactions among heterogeneous agents-each with their own preferences, and potentially strategic behavior.Analyzing the outcomes that emerge from the interactions of decentralized agents-and understanding their implications for overall welfare-is a central and challenging problem at the intersection of computer science, economics, and operations research. These challenges become especially pronounced in algorithmically powered environments, where systems are non-stationary, and agents have limited information, adapting their behavior in real-time. In such dynamic settings, traditional equilibrium and welfare analysis is often not applicable, opening the way to the definition and analysis of new equilibrium and welfare metrics. In the context of preference aggregation, traditional aggregation models have assumed that disagreements between users' preferences carry equal intensity regardless of the position, and that all alternatives are interchangeable. These assumptions may be analytically convenient, but they are increasingly implausible in modern applications such as recommender systems, where users' preferences are aggregated to provide new recommendations. For instance, in recommender systems, users pay far more attention to top-ranked suggestions than to lower ones-consider for example the difference between the first and tenth movie on a streaming platform-and alternatives often differ in intrinsic characteristics, such as genre, brand, or quality tier. Understanding how to measure differences in users' preferences accounting for these features becomes essential to devise effective rank aggregation mechanisms.The first chapter of my thesis is based on my work Equilibria with Dynamic Benchmarks in NonStationary Multi-Agent Systems,co-authored with Yonatan Gur and Bar Light. In this paper, we formulate and study a general time-varying multi-agent system where players repeatedly compete under incomplete information. Our work is motivated by scenarios commonly observed in online advertising and retail marketplaces, where agents and platform designers optimize algorithmic decision-making in dynamic competitive settings. In these systems, no-regret algorithms that provide guarantees relative to staticbenchmarks can perform poorly and the distributions of play that emerge from their interaction do not correspond anymore to static solution concepts such as coarse correlated equilibria. Instead, we analyze the interaction of dynamic benchmarkconsistent policies that have performance guarantees relative to dynamicsequences of actions, and through a novel tracking errornotion we delineate when their empirical joint distribution of play can approximate an evolving sequence of static equilibria. In systems that change sufficiently slowly (sub-linearly in the horizon length), we show that the resulting distributions of play approximate the sequence of coarse correlated equilibria, and apply this result to establish improved welfare bounds for smooth games. On a similar vein, we formulate internal dynamic benchmark consistent policies and establish that they approximate sequences of correlated equilibria. Our findings therefore suggest that, in a broad range of multi-agent systems where non-stationarity is prevalent, algorithms designed to compete with dynamic benchmarks can improve both individual and welfare guarantees, and their emerging dynamics approximate a sequence of static equilibrium outcomes.
■590 ▼aSchool code: 0212.
■650 4▼aMotivation
■650 4▼aInteger programming
■650 4▼aDecision making
■650 4▼aBenchmarks
■650 4▼aPrices
■650 4▼aComputer science
■690 ▼a0984
■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=T17359280▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


