서브메뉴
검색
Learning and Optimization Over Robust Networked Systems
Learning and Optimization Over Robust Networked Systems
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105157
- ISBN
- 9798293877386
- DDC
- 621.3
- 서명/저자
- Learning and Optimization Over Robust Networked Systems
- 발행사항
- [Sl] : Carnegie Mellon University, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 197 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-03, Section: B.
- 주기사항
- Advisor: Joe-Wong, Carlee;Yagan, Osman.
- 학위논문주기
- Thesis (Ph.D.)--Carnegie Mellon University, 2025.
- 초록/해제
- 요약As networked systems grow in scale and complexity, they become increasingly vulnerable to failures, congestion, and inefficiencies that may lead to system-wide breakdowns. To address these challenges, this work develops adaptive frameworks, algorithmic solutions, and theoretical analyzes that improve robustness, scalability, learning, and personalization. We investigate optimization and learning in robust networked systems across three interconnected domains: interdependent network resilience, next-generation transportation systems, and decentralized federated learning.First, we examine cascading failures in interdependent networks. A novel dynamic coupling strategy is introduced, which adaptively adjusts load redistribution in real time. This strategy is mathematically characterized, computationally efficient, and empirically validated to improve system survival rates compared to static approaches, significantly increasing the critical attack threshold, the minimum initial disruption required to trigger total collapse.Second, the research extends resilience analysis to transportation networks under mixed autonomy, where autonomous vehicles (AVs) coexist with human-driven vehicles (HVs). Using a game-theoretic framework, the thesis derives equilibrium strategies for AV routing and flow re-balancing that mitigate congestion propagation. To capture the dynamic and decentralized nature of real traffic, a multi-agent reinforcement learning (MARL) framework is further developed for fleet management, enabling scalable coordination of AVs that improves throughput and network resilience without centralized control.Third, we address the pressing scalability and personalization challenges in federated learning, particularly in decentralized settings where no central server coordinates training and client data are highly heterogeneous. Existing methods often struggle with non-IID distributions, communication bottlenecks, and the need for models that adapt to diverse users. To overcome these limitations, we propose FedSPD (Federated learning with Soft-clustering Personalized Decentralized), which enhances decentralized federated learning by enabling clients to form flexible, soft clusters that balance global collaboration with personalized adaptation. This design directly tackles data heterogeneity and connectivity constraints, allowing effective learning even in low-resource or sparsely connected networks.Together, these contributions establish a systematic approach for building intelligent, adaptive, and failure-resistant infrastructures. The results offer both theoretical insights and practical frameworks, providing a foundation for the development of secure, scalable, and trustworthy networked systems in an increasingly interconnected world.
- 일반주제명
- Computer engineering
- 일반주제명
- Computer science
- 일반주제명
- Electrical engineering
- 키워드
- AV routing
- 기타저자
- Carnegie Mellon University Electrical and Computer Engineering
- 기본자료저록
- Dissertations Abstracts International. 87-03B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017359681
■00520260202105157
■006m o d
■007cr#unu||||||||
■020 ▼a9798293877386
■035 ▼a(MiAaPQ)AAI32243567
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a621.3
■1001 ▼aLin, I-Cheng Delphi.▼0(orcid)0000-0002-5306-3262
■24510▼aLearning and Optimization Over Robust Networked Systems
■260 ▼a[Sl]▼bCarnegie Mellon University▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a197 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-03, Section: B.
■500 ▼aAdvisor: Joe-Wong, Carlee;Yagan, Osman.
■5021 ▼aThesis (Ph.D.)--Carnegie Mellon University, 2025.
■520 ▼aAs networked systems grow in scale and complexity, they become increasingly vulnerable to failures, congestion, and inefficiencies that may lead to system-wide breakdowns. To address these challenges, this work develops adaptive frameworks, algorithmic solutions, and theoretical analyzes that improve robustness, scalability, learning, and personalization. We investigate optimization and learning in robust networked systems across three interconnected domains: interdependent network resilience, next-generation transportation systems, and decentralized federated learning.First, we examine cascading failures in interdependent networks. A novel dynamic coupling strategy is introduced, which adaptively adjusts load redistribution in real time. This strategy is mathematically characterized, computationally efficient, and empirically validated to improve system survival rates compared to static approaches, significantly increasing the critical attack threshold, the minimum initial disruption required to trigger total collapse.Second, the research extends resilience analysis to transportation networks under mixed autonomy, where autonomous vehicles (AVs) coexist with human-driven vehicles (HVs). Using a game-theoretic framework, the thesis derives equilibrium strategies for AV routing and flow re-balancing that mitigate congestion propagation. To capture the dynamic and decentralized nature of real traffic, a multi-agent reinforcement learning (MARL) framework is further developed for fleet management, enabling scalable coordination of AVs that improves throughput and network resilience without centralized control.Third, we address the pressing scalability and personalization challenges in federated learning, particularly in decentralized settings where no central server coordinates training and client data are highly heterogeneous. Existing methods often struggle with non-IID distributions, communication bottlenecks, and the need for models that adapt to diverse users. To overcome these limitations, we propose FedSPD (Federated learning with Soft-clustering Personalized Decentralized), which enhances decentralized federated learning by enabling clients to form flexible, soft clusters that balance global collaboration with personalized adaptation. This design directly tackles data heterogeneity and connectivity constraints, allowing effective learning even in low-resource or sparsely connected networks.Together, these contributions establish a systematic approach for building intelligent, adaptive, and failure-resistant infrastructures. The results offer both theoretical insights and practical frameworks, providing a foundation for the development of secure, scalable, and trustworthy networked systems in an increasingly interconnected world.
■590 ▼aSchool code: 0041.
■650 4▼aComputer engineering
■650 4▼aComputer science
■650 4▼aElectrical engineering
■653 ▼aRobust networked systems
■653 ▼aSystem-wide breakdowns
■653 ▼aHuman-driven vehicles
■653 ▼aMulti-agent reinforcement learning
■653 ▼aAV routing
■690 ▼a0464
■690 ▼a0800
■690 ▼a0984
■690 ▼a0544
■71020▼aCarnegie Mellon University▼bElectrical and Computer Engineering.
■7730 ▼tDissertations Abstracts International▼g87-03B.
■790 ▼a0041
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17359681▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
