서브메뉴
검색
Spatio-Temporal Representation Learning: Applications to Manufacturing Planning and Pedestrian Crowd Analysis
Spatio-Temporal Representation Learning: Applications to Manufacturing Planning and Pedestrian Crowd Analysis
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211152115
- ISBN
- 9798384340195
- DDC
- 796.34
- 서명/저자
- Spatio-Temporal Representation Learning: Applications to Manufacturing Planning and Pedestrian Crowd Analysis
- 발행사항
- [Sl] : Stanford University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 134 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-03, Section: B.
- 주기사항
- Advisor: Law, Kincho;Fruchter, Renate;Lepech, Michael.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2024.
- 초록/해제
- 요약The recent availability of data and advancements in computational tools have opened up new possibilities and challenges in urban planning and city management. Data-driven spatio-temporal models are continuously being developed to study complex activities within cities, such as supply chain (the flow of goods) and urban mobility (the flow of people). This thesis focuses on applying spatio-temporal representation learning techniques to two main application domains: manufacturing planning and pedestrian crowd analysis. In the manufacturing planning domain, we address the interruptive swap-allowed job shop scheduling problem (ISBJSSP), using graph neural networks (GNNs) and reinforcement learning (RL) to generate adaptive scheduling policies. We develop a simulator to model real-world manufacturing constraints and validate our approach using benchmark instances.In the pedestrian crowd analysis domain, we propose a computer vision-based procedure to generate crowd data from raw surveillance footage and spatial connectivity priors, introducing crowd mobility graphs (CMGraphs) to represent aggregated crowd flow. Subsequently, we formulate the crowd flow forecasting problem to predict future crowd states and design a neural network framework (STEN) that utilizes spatial and temporal information for this task. We work with both public single-camera video data (Grand Central Station dataset) and real-world collected multi-camera video data (Campus Crowd dataset) to validate our approaches and provide resources for future research.The main contributions of this thesis are threefold: (1) spatio-temporal data acquisition from three sources: simulated, public single-camera, and real-world multi-camera; (2) problem definition and spatio-temporal graph representation of two problems: ISBJSSP and crowd flow forecasting; and (3) two deep learning model implementation to generate solutions for the defined problems under architectural frameworks: GNN-RL and STEN.
- 일반주제명
- Table tennis
- 일반주제명
- Scheduling
- 일반주제명
- Families & family life
- 일반주제명
- Forecasting
- 일반주제명
- Graph representations
- 일반주제명
- Benchmarks
- 일반주제명
- Industrial engineering
- 일반주제명
- Recreation
- 기타저자
- Stanford University.
- 기본자료저록
- Dissertations Abstracts International. 86-03B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017162946
■00520250211152115
■006m o d
■007cr#unu||||||||
■020 ▼a9798384340195
■035 ▼a(MiAaPQ)AAI31460291
■035 ▼a(MiAaPQ)Stanfordhz688jc2430
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a796.34
■1001 ▼aWong, Vivian W.H.
■24510▼aSpatio-Temporal Representation Learning: Applications to Manufacturing Planning and Pedestrian Crowd Analysis
■260 ▼a[Sl]▼bStanford University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a134 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-03, Section: B.
■500 ▼aAdvisor: Law, Kincho;Fruchter, Renate;Lepech, Michael.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2024.
■520 ▼aThe recent availability of data and advancements in computational tools have opened up new possibilities and challenges in urban planning and city management. Data-driven spatio-temporal models are continuously being developed to study complex activities within cities, such as supply chain (the flow of goods) and urban mobility (the flow of people). This thesis focuses on applying spatio-temporal representation learning techniques to two main application domains: manufacturing planning and pedestrian crowd analysis. In the manufacturing planning domain, we address the interruptive swap-allowed job shop scheduling problem (ISBJSSP), using graph neural networks (GNNs) and reinforcement learning (RL) to generate adaptive scheduling policies. We develop a simulator to model real-world manufacturing constraints and validate our approach using benchmark instances.In the pedestrian crowd analysis domain, we propose a computer vision-based procedure to generate crowd data from raw surveillance footage and spatial connectivity priors, introducing crowd mobility graphs (CMGraphs) to represent aggregated crowd flow. Subsequently, we formulate the crowd flow forecasting problem to predict future crowd states and design a neural network framework (STEN) that utilizes spatial and temporal information for this task. We work with both public single-camera video data (Grand Central Station dataset) and real-world collected multi-camera video data (Campus Crowd dataset) to validate our approaches and provide resources for future research.The main contributions of this thesis are threefold: (1) spatio-temporal data acquisition from three sources: simulated, public single-camera, and real-world multi-camera; (2) problem definition and spatio-temporal graph representation of two problems: ISBJSSP and crowd flow forecasting; and (3) two deep learning model implementation to generate solutions for the defined problems under architectural frameworks: GNN-RL and STEN.
■590 ▼aSchool code: 0212.
■650 4▼aTable tennis
■650 4▼aScheduling
■650 4▼aFamilies & family life
■650 4▼aForecasting
■650 4▼aGraph representations
■650 4▼aBenchmarks
■650 4▼aIndividual & family studies
■650 4▼aIndustrial engineering
■650 4▼aRecreation
■690 ▼a0628
■690 ▼a0546
■690 ▼a0814
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g86-03B.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17162946▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
