서브메뉴
검색
Understanding and Mitigating Bias in Algorithms, Data, and Society
Understanding and Mitigating Bias in Algorithms, Data, and Society
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105545
- ISBN
- 9798265400840
- DDC
- 658
- 저자명
- Zhao, Zhanzhan.
- 서명/저자
- Understanding and Mitigating Bias in Algorithms, Data, and Society
- 발행사항
- [Sl] : Georgia Institute of Technology, 2023
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2023
- 형태사항
- 225 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-05, Section: A.
- 주기사항
- Advisor: Randall, Dana.
- 학위논문주기
- Thesis (Ph.D.)--Georgia Institute of Technology, 2023.
- 초록/해제
- 요약Societal biases are systemic prejudices embedded in individuals' and institutions' collective beliefs, attitudes, norms, and behaviors. These biases, such as racial bias, ideological bias, and partisan bias, no matter explicit or implicit, are often deeply ingrained and can be reinforced by social, cultural, and structural factors. While computer scientists have made significant strides in quantifying biased patterns and developing mitigation strategies for relevant computer algorithms and internet applications, exploring the root causes of societal biases remain a challenging problem. Due to the complex nature and unpredictability of the emergent societal behaviors with limited data availability, many powerful algorithms and data analysis tools in computer science face limitations when addressing these issues.This thesis focuses on enhancing algorithms and data analytics tools for addressing societal biases. We take a multi-lens approach, combining algorithmic modeling, data analysis, and qualitative theory to gain comprehensive insights into three specific themes: the persistent residential segregation arising from racial homophily, ideological polarization exacerbated by reinforcements of beliefs, and non-responsiveness in political elections boosted by partisan redistricting policies. By integrating these different research methods, we aim to develop a deeper understanding of these equity-related social problems and propose system-level mitigation interventions.Drawing tools from probability and theoretical computer science, we first propose a rigorous approach to understand the possible causal mechanisms behind residential segregation. We proved how carefully architected placement of incentives, such as urban amenities, might worsen or mitigate segregation. In collaboration with regional planning scholars, we confirmed and enriched our theoretical findings using machine learning analysis and detected several distinct biased public amenity distribution patterns accompanying various severity of segregation across U.S. cities.Just as unintentional placements of public amenities may exacerbate segregation, we also use agent-based modeling to show the designs of highly personalized recommender algorithms to stimulate users' interests can contribute to the escalation of extreme polarization. Through our simulations, we demonstrate the effectiveness of our proposed novel designs of local recommendations that adaptively diversify and consider individual tolerance levels on reversing the polarization trend, even in the presence of extreme influencers and social pressure, highlighting the significance of tailored interventions that carefully leverage the power of diversity to promote inclusive dialogue and bridge ideological divides.Besides proposing system-level interventions to mitigate racial and ideological biases, we also make advances on generating fair baselines for evaluating potential partisan biases in political redistricting policies. By proposing a novel batched parallel tempering protocol with a proof of improvement in computational efficiency, we are able to sample policies from law requirements captured by probability distributions. Through the first extensive computational study on Georgia's congressional and state districting plans, we show that the enacted plans are much less responsive to changing electoral opinions than one should expect, even when accounting for both Voting Rights Act considerations and traditional redistricting criteria, and we localize areas that are most likely responsible.
- 일반주제명
- Behavior
- 일반주제명
- Segregation
- 일반주제명
- Computer science
- 일반주제명
- Ideology
- 일반주제명
- Inclusion
- 일반주제명
- Redistricting
- 일반주제명
- Society
- 일반주제명
- Prejudice
- 일반주제명
- Voting Rights Act
- 일반주제명
- Spatial analysis
- 일반주제명
- Election results
- 일반주제명
- State elections
- 일반주제명
- Governors
- 일반주제명
- Presidential elections
- 일반주제명
- Race
- 일반주제명
- Political science
- 일반주제명
- Sociology
- 기본자료저록
- Dissertations Abstracts International. 87-05A.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2023 us c eng d■001000017360548
■00520260202105545
■006m o d
■007cr#unu||||||||
■020 ▼a9798265400840
■035 ▼a(MiAaPQ)AAI32315542
■035 ▼a(MiAaPQ)GeorgiaTech75587
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a658
■1001 ▼aZhao, Zhanzhan.
■24510▼aUnderstanding and Mitigating Bias in Algorithms, Data, and Society
■260 ▼a[Sl]▼bGeorgia Institute of Technology▼c2023
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2023
■300 ▼a225 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-05, Section: A.
■500 ▼aAdvisor: Randall, Dana.
■5021 ▼aThesis (Ph.D.)--Georgia Institute of Technology, 2023.
■520 ▼aSocietal biases are systemic prejudices embedded in individuals' and institutions' collective beliefs, attitudes, norms, and behaviors. These biases, such as racial bias, ideological bias, and partisan bias, no matter explicit or implicit, are often deeply ingrained and can be reinforced by social, cultural, and structural factors. While computer scientists have made significant strides in quantifying biased patterns and developing mitigation strategies for relevant computer algorithms and internet applications, exploring the root causes of societal biases remain a challenging problem. Due to the complex nature and unpredictability of the emergent societal behaviors with limited data availability, many powerful algorithms and data analysis tools in computer science face limitations when addressing these issues.This thesis focuses on enhancing algorithms and data analytics tools for addressing societal biases. We take a multi-lens approach, combining algorithmic modeling, data analysis, and qualitative theory to gain comprehensive insights into three specific themes: the persistent residential segregation arising from racial homophily, ideological polarization exacerbated by reinforcements of beliefs, and non-responsiveness in political elections boosted by partisan redistricting policies. By integrating these different research methods, we aim to develop a deeper understanding of these equity-related social problems and propose system-level mitigation interventions.Drawing tools from probability and theoretical computer science, we first propose a rigorous approach to understand the possible causal mechanisms behind residential segregation. We proved how carefully architected placement of incentives, such as urban amenities, might worsen or mitigate segregation. In collaboration with regional planning scholars, we confirmed and enriched our theoretical findings using machine learning analysis and detected several distinct biased public amenity distribution patterns accompanying various severity of segregation across U.S. cities.Just as unintentional placements of public amenities may exacerbate segregation, we also use agent-based modeling to show the designs of highly personalized recommender algorithms to stimulate users' interests can contribute to the escalation of extreme polarization. Through our simulations, we demonstrate the effectiveness of our proposed novel designs of local recommendations that adaptively diversify and consider individual tolerance levels on reversing the polarization trend, even in the presence of extreme influencers and social pressure, highlighting the significance of tailored interventions that carefully leverage the power of diversity to promote inclusive dialogue and bridge ideological divides.Besides proposing system-level interventions to mitigate racial and ideological biases, we also make advances on generating fair baselines for evaluating potential partisan biases in political redistricting policies. By proposing a novel batched parallel tempering protocol with a proof of improvement in computational efficiency, we are able to sample policies from law requirements captured by probability distributions. Through the first extensive computational study on Georgia's congressional and state districting plans, we show that the enacted plans are much less responsive to changing electoral opinions than one should expect, even when accounting for both Voting Rights Act considerations and traditional redistricting criteria, and we localize areas that are most likely responsible.
■590 ▼aSchool code: 0078.
■650 4▼aBehavior
■650 4▼aSegregation
■650 4▼aComputer science
■650 4▼aIdeology
■650 4▼aInclusion
■650 4▼aRedistricting
■650 4▼aSociety
■650 4▼aPrejudice
■650 4▼aVoting Rights Act
■650 4▼aSpatial analysis
■650 4▼aElection results
■650 4▼aState elections
■650 4▼aMulticulturalism & pluralism
■650 4▼aGovernors
■650 4▼aPresidential elections
■650 4▼aRace
■650 4▼aPolitical science
■650 4▼aSociology
■690 ▼a0984
■690 ▼a0615
■690 ▼a0626
■71020▼aGeorgia Institute of Technology.
■7730 ▼tDissertations Abstracts International▼g87-05A.
■790 ▼a0078
■791 ▼aPh.D.
■792 ▼a2023
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360548▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


