서브메뉴
검색
Human-Centered Machine Learning in Operations Management
Human-Centered Machine Learning in Operations Management
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105100
- ISBN
- 9798293893492
- DDC
- 004
- 저자명
- Jiang, Shunan.
- 서명/저자
- Human-Centered Machine Learning in Operations Management
- 발행사항
- [Sl] : University of California, Berkeley, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 174 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-04, Section: B.
- 주기사항
- Advisor: Shen, Zuo-Jun Max;Sinchaisri, Park.
- 학위논문주기
- Thesis (Ph.D.)--University of California, Berkeley, 2025.
- 초록/해제
- 요약This dissertation investigates the operational challenges and opportunities at the intersection of machine learning, algorithmic design, and human behavior in modern digital platforms. As services in human-in-the-loop systems increasingly rely on AI-driven systems, a critical tension emerges between platform-level optimization and the agency of autonomous human participants. This work addresses this tension through a series of empirical studies, theoretical models, and novel algorithms, adopting a human-centered perspective to improve system design and performance.First, we conduct a large-scale empirical study of a retail delivery platform to understand how gig workers learn and adapt to algorithmic recommendations. Our analysis reveals that workers evolve sophisticated, personalized strategies over time, transitioning from initial reliance on platform guidance to a dynamic co-adaptation with the system.Second, motivated by these empirical findings, we formalize the problem of assigning tasks to a workforce of independent, learning agents. We introduce a novel two-phase bandit model that captures worker agency and endogenous skill development. We prove that the optimal policy is intractable and propose a practical algorithm, Combinatorial UCB (C-UCB), which we show achieves sublinear regret against a strong greedy oracle.Third, we shift our focus to user engagement, examining the impact of AI-driven matchmaking in an online gaming context. We find that AI's influence is a "double-edged sword": while moderate AI involvement can enhance short-term engagement, particularly for novice players, excessive exposure is negatively associated with long-term retention. This highlights the need for careful calibration of AI to avoid undermining user trust and intrinsic motivation.Finally, we address the classic operations problem of inventory management in the context of on-demand vehicle sharing networks with censored demand. We model the system as an infinite horizon Markov Decision Process and develop an online learning algorithm, Learning While Repositioning (LWR), that learns an effective repositioning policy despite incomplete information. We establish a sublinear regret bound for our algorithm, demonstrating its efficacy in a high-dimensional, dynamic setting.Collectively, these studies contribute a multi-faceted understanding of human-algorithm interaction in operations management. By bridging empirical analysis with theoretical rigor, this dissertation provides new models and actionable insights for designing more efficient, adaptive, and human-responsive platforms.
- 일반주제명
- Computer science
- 키워드
- Machine learning
- 키워드
- Novel algorithms
- 기타저자
- University of California, Berkeley Industrial Engineering & Operations Research
- 기본자료저록
- Dissertations Abstracts International. 87-04B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017359315
■00520260202105100
■006m o d
■007cr#unu||||||||
■020 ▼a9798293893492
■035 ▼a(MiAaPQ)AAI32235917
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a004
■1001 ▼aJiang, Shunan.
■24510▼aHuman-Centered Machine Learning in Operations Management
■260 ▼a[Sl]▼bUniversity of California, Berkeley▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a174 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-04, Section: B.
■500 ▼aAdvisor: Shen, Zuo-Jun Max;Sinchaisri, Park.
■5021 ▼aThesis (Ph.D.)--University of California, Berkeley, 2025.
■520 ▼aThis dissertation investigates the operational challenges and opportunities at the intersection of machine learning, algorithmic design, and human behavior in modern digital platforms. As services in human-in-the-loop systems increasingly rely on AI-driven systems, a critical tension emerges between platform-level optimization and the agency of autonomous human participants. This work addresses this tension through a series of empirical studies, theoretical models, and novel algorithms, adopting a human-centered perspective to improve system design and performance.First, we conduct a large-scale empirical study of a retail delivery platform to understand how gig workers learn and adapt to algorithmic recommendations. Our analysis reveals that workers evolve sophisticated, personalized strategies over time, transitioning from initial reliance on platform guidance to a dynamic co-adaptation with the system.Second, motivated by these empirical findings, we formalize the problem of assigning tasks to a workforce of independent, learning agents. We introduce a novel two-phase bandit model that captures worker agency and endogenous skill development. We prove that the optimal policy is intractable and propose a practical algorithm, Combinatorial UCB (C-UCB), which we show achieves sublinear regret against a strong greedy oracle.Third, we shift our focus to user engagement, examining the impact of AI-driven matchmaking in an online gaming context. We find that AI's influence is a "double-edged sword": while moderate AI involvement can enhance short-term engagement, particularly for novice players, excessive exposure is negatively associated with long-term retention. This highlights the need for careful calibration of AI to avoid undermining user trust and intrinsic motivation.Finally, we address the classic operations problem of inventory management in the context of on-demand vehicle sharing networks with censored demand. We model the system as an infinite horizon Markov Decision Process and develop an online learning algorithm, Learning While Repositioning (LWR), that learns an effective repositioning policy despite incomplete information. We establish a sublinear regret bound for our algorithm, demonstrating its efficacy in a high-dimensional, dynamic setting.Collectively, these studies contribute a multi-faceted understanding of human-algorithm interaction in operations management. By bridging empirical analysis with theoretical rigor, this dissertation provides new models and actionable insights for designing more efficient, adaptive, and human-responsive platforms.
■590 ▼aSchool code: 0028.
■650 4▼aComputer science
■653 ▼aMachine learning
■653 ▼aLearning While Repositioning
■653 ▼aNovel algorithms
■653 ▼aHuman-algorithm interaction
■653 ▼aOnline learning algorithm
■690 ▼a0796
■690 ▼a0984
■690 ▼a0800
■71020▼aUniversity of California, Berkeley▼bIndustrial Engineering & Operations Research.
■7730 ▼tDissertations Abstracts International▼g87-04B.
■790 ▼a0028
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17359315▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
