서브메뉴
검색
Granular Feedback Merits Sophisticated Aggregation
Granular Feedback Merits Sophisticated Aggregation
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105626
- ISBN
- 9798265428820
- DDC
- 519.7
- 저자명
- Kagrecha, Anmol.
- 서명/저자
- Granular Feedback Merits Sophisticated Aggregation
- 발행사항
- [Sl] : Stanford University, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 61 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-05, Section: B.
- 주기사항
- Advisor: Van Roy, Benjamin.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2025.
- 초록/해제
- 요약Human feedback is increasingly used across diverse applications like training AI models, developing recommender systems, and measuring public opinion - with granular feedback often being preferred over binary feedback for its greater informativeness. While it is easy to accurately estimate a population's distribution of feedback given feedback from a large number of individuals, cost constraints typically necessitate using smaller groups. A simple method to approximate the population distribution is regularized averaging: compute the empirical distribution and regularize it toward a prior. Can we do better? As we will discuss, the answer to this question depends on feedback granularity.Suppose one wants to predict a population's distribution of feedback using feedback from a limited number of individuals. We show that, as feedback granularity increases, one can substantially improve upon predictions of regularized averaging by combining individuals' feedback in ways more sophisticated than regularized averaging.Our empirical analysis using questions on social attitudes confirms this pattern. In particular, with binary feedback, sophistication barely reduces the number of individuals required to attain a fixed level of performance. By contrast, with five-point feedback, sophisticated methods match the performance of regularized averaging with about half as many individuals.Besides discussing how the advantage of sophisticated aggregation methods varies with granularity, we also discuss two results. First, we discuss how sophisticated aggregation methods can improve a typical pipeline for training large language models. Second, we demonstrate experimentally that if one has interacted sufficiently many times with a set of individuals, then one could collect feedback from only a subset of these individuals while maintaining equivalent performance. Combined with our main result that sophisticated aggregation methods provide a larger advantage for granular feedback, this subset selection capability implies that the joint use of sophisticated aggregation and selection can substantially outperform the combination of regularized averaging and random selection when feedback is granular. A major part of this dissertation is adapted from a recent preprint (Kagrecha et al., 2025) authored by Anmol Kagrecha, Henrik Marklund, Potsawee Manakul, Richard Zeckhauser, and Benjamin Van Roy.
- 일반주제명
- Optimization techniques
- 일반주제명
- Families & family life
- 일반주제명
- Attitudes
- 일반주제명
- Abortion
- 일반주제명
- Crowdsourcing
- 일반주제명
- Web studies
- 일반주제명
- Obstetrics
- 기타저자
- Stanford University.
- 기본자료저록
- Dissertations Abstracts International. 87-05B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017360839
■00520260202105626
■006m o d
■007cr#unu||||||||
■020 ▼a9798265428820
■035 ▼a(MiAaPQ)AAI32316560
■035 ▼a(MiAaPQ)Stanfordpd147dc7682
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a519.7
■1001 ▼aKagrecha, Anmol.
■24510▼aGranular Feedback Merits Sophisticated Aggregation
■260 ▼a[Sl]▼bStanford University▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a61 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-05, Section: B.
■500 ▼aAdvisor: Van Roy, Benjamin.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2025.
■520 ▼aHuman feedback is increasingly used across diverse applications like training AI models, developing recommender systems, and measuring public opinion - with granular feedback often being preferred over binary feedback for its greater informativeness. While it is easy to accurately estimate a population's distribution of feedback given feedback from a large number of individuals, cost constraints typically necessitate using smaller groups. A simple method to approximate the population distribution is regularized averaging: compute the empirical distribution and regularize it toward a prior. Can we do better? As we will discuss, the answer to this question depends on feedback granularity.Suppose one wants to predict a population's distribution of feedback using feedback from a limited number of individuals. We show that, as feedback granularity increases, one can substantially improve upon predictions of regularized averaging by combining individuals' feedback in ways more sophisticated than regularized averaging.Our empirical analysis using questions on social attitudes confirms this pattern. In particular, with binary feedback, sophistication barely reduces the number of individuals required to attain a fixed level of performance. By contrast, with five-point feedback, sophisticated methods match the performance of regularized averaging with about half as many individuals.Besides discussing how the advantage of sophisticated aggregation methods varies with granularity, we also discuss two results. First, we discuss how sophisticated aggregation methods can improve a typical pipeline for training large language models. Second, we demonstrate experimentally that if one has interacted sufficiently many times with a set of individuals, then one could collect feedback from only a subset of these individuals while maintaining equivalent performance. Combined with our main result that sophisticated aggregation methods provide a larger advantage for granular feedback, this subset selection capability implies that the joint use of sophisticated aggregation and selection can substantially outperform the combination of regularized averaging and random selection when feedback is granular. A major part of this dissertation is adapted from a recent preprint (Kagrecha et al., 2025) authored by Anmol Kagrecha, Henrik Marklund, Potsawee Manakul, Richard Zeckhauser, and Benjamin Van Roy.
■590 ▼aSchool code: 0212.
■650 4▼aOptimization techniques
■650 4▼aFamilies & family life
■650 4▼aAttitudes
■650 4▼aAbortion
■650 4▼aCrowdsourcing
■650 4▼aIndividual & family studies
■650 4▼aWeb studies
■650 4▼aObstetrics
■690 ▼a0628
■690 ▼a0646
■690 ▼a0380
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g87-05B.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360839▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


