서브메뉴
검색
Inference and Information in Network Structure
Inference and Information in Network Structure
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105218
- ISBN
- 9798291565902
- DDC
- 310
- 서명/저자
- Inference and Information in Network Structure
- 발행사항
- [Sl] : University of Michigan, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 262 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-03, Section: B.
- 주기사항
- Advisor: Newman, Mark.
- 학위논문주기
- Thesis (Ph.D.)--University of Michigan, 2025.
- 초록/해제
- 요약Many systems across science can be meaningfully represented as networks of simple interactions. Within networks of metabolic pathways, transportation links, neuronal connections, outcomes of sports matches, and social ties, the patterns and directions of data often exhibit collective behavior. In this thesis we focus on characterizing two common structural features of these networks: group structure and hierarchy.In real-world networks we often find communities - groups of nodes which interact with each other more frequently than with nodes outside their group. Common examples include friend groups, functional neuronal groups, or ecological niches. Despite their ubiquity, network data does not often come to us already labeled with this group structure; it must be algorithmically inferred from the network alone. In this work, we describe a number of refinements of existing community detection models which allow us to: discover more and smaller groups, directly measure the in-group preference within the system, and measure the variation in connections within groups.When the ground truth community structure of a network is available, the outputs of these community detection algorithms are often compared against that truth in an information theoretic manner. We discuss improvements to this framework that address a bias towards algorithms which find an excessive number of communities and better quantify relevant information costs. We further demonstrate how the conclusions drawn depend on the form of this measure in extensive tests on synthetic networks. Similarly, when we observe directed relationships such as dominance interactions among animals or humans, the directions of faculty hiring among universities, or wins and losses in games and sports, hierarchies routinely emerge. In this work we draw an analogy between fermion energy and competitor rank in these hierarchies to define a notion of the collective "temperature" of a hierarchy which we may then measure. This parameter then also indicates the strictness of the hierarchy, or equivalently the number of distinct levels of play. We find a good deal of variation in this measured temperature: sports rankings tend to be hot and unpredictable, animal hierarchies are cold and rigid, while human social hierarchies are lukewarm - somewhere in the middle.Taken together, our work has not only enabled new insights about longstanding problems of network science but also offered a path to understand and measure their nature.
- 일반주제명
- Statistics
- 일반주제명
- Physics
- 일반주제명
- Social structure
- 일반주제명
- Systems science
- 키워드
- Networks
- 키워드
- Hierarchy
- 기타저자
- University of Michigan Physics
- 기본자료저록
- Dissertations Abstracts International. 87-03B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017359810
■00520260202105218
■006m o d
■007cr#unu||||||||
■020 ▼a9798291565902
■035 ▼a(MiAaPQ)AAI32271778
■035 ▼a(MiAaPQ)umichrackham006404
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a310
■1001 ▼aJerdee, Maximilian.
■24510▼aInference and Information in Network Structure
■260 ▼a[Sl]▼bUniversity of Michigan▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a262 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-03, Section: B.
■500 ▼aAdvisor: Newman, Mark.
■5021 ▼aThesis (Ph.D.)--University of Michigan, 2025.
■520 ▼aMany systems across science can be meaningfully represented as networks of simple interactions. Within networks of metabolic pathways, transportation links, neuronal connections, outcomes of sports matches, and social ties, the patterns and directions of data often exhibit collective behavior. In this thesis we focus on characterizing two common structural features of these networks: group structure and hierarchy.In real-world networks we often find communities - groups of nodes which interact with each other more frequently than with nodes outside their group. Common examples include friend groups, functional neuronal groups, or ecological niches. Despite their ubiquity, network data does not often come to us already labeled with this group structure; it must be algorithmically inferred from the network alone. In this work, we describe a number of refinements of existing community detection models which allow us to: discover more and smaller groups, directly measure the in-group preference within the system, and measure the variation in connections within groups.When the ground truth community structure of a network is available, the outputs of these community detection algorithms are often compared against that truth in an information theoretic manner. We discuss improvements to this framework that address a bias towards algorithms which find an excessive number of communities and better quantify relevant information costs. We further demonstrate how the conclusions drawn depend on the form of this measure in extensive tests on synthetic networks. Similarly, when we observe directed relationships such as dominance interactions among animals or humans, the directions of faculty hiring among universities, or wins and losses in games and sports, hierarchies routinely emerge. In this work we draw an analogy between fermion energy and competitor rank in these hierarchies to define a notion of the collective "temperature" of a hierarchy which we may then measure. This parameter then also indicates the strictness of the hierarchy, or equivalently the number of distinct levels of play. We find a good deal of variation in this measured temperature: sports rankings tend to be hot and unpredictable, animal hierarchies are cold and rigid, while human social hierarchies are lukewarm - somewhere in the middle.Taken together, our work has not only enabled new insights about longstanding problems of network science but also offered a path to understand and measure their nature.
■590 ▼aSchool code: 0127.
■650 4▼aStatistics
■650 4▼aPhysics
■650 4▼aSocial structure
■650 4▼aSystems science
■653 ▼aNetworks
■653 ▼aStatistical inference
■653 ▼aCommunity structure
■653 ▼aHierarchy
■653 ▼aInformation theory
■690 ▼a0605
■690 ▼a0463
■690 ▼a0700
■690 ▼a0790
■71020▼aUniversity of Michigan▼bPhysics.
■7730 ▼tDissertations Abstracts International▼g87-03B.
■790 ▼a0127
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17359810▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
