서브메뉴
검색
Statistical Methods for Efficient and Trustworthy Machine Learning
Statistical Methods for Efficient and Trustworthy Machine Learning
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211153055
- ISBN
- 9798346390039
- DDC
- 300
- 저자명
- Isik, Berivan.
- 서명/저자
- Statistical Methods for Efficient and Trustworthy Machine Learning
- 발행사항
- [Sl] : Stanford University, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 268 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-05, Section: B.
- 주기사항
- Advisor: Weissman, Tsachy.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2024.
- 초록/해제
- 요약This thesis addresses key challenges in machine learning, with a focus on efficiency and trustworthiness through statistical tools. It explores emerging topics in the era of large models and big data, including model compression, federated learning, and data privacy.New methods for model and training data compression are introduced, leveraging rate-distortion theory and novel coding techniques designed for the specific demands of machine learning tasks. Additionally, new federated learning frameworks are proposed to reduce communication costs while maintaining or even enhancing accuracy and inference efficiency. These frameworks employ strategies such as model update sparsification, exploiting overparameterization in modern models, and utilizing side information to achieve substantial bitrate reductions without sacrificing performance.The research advances the understanding of compression-privacy-utility tradeoffs in training data processing and distributed mean estimation, offering new optimal algorithms that effectively balance these tradeoffs. These contributions lay a robust theoretical foundation for enhancing data privacy while reducing communication and storage costs, all without compromising utility.Overall, this work advances machine learning systems by proposing solutions that address the unique conditions, limitations, and flexibilities of these models, ultimately enhancing efficiency and trustworthiness across multiple domains.
- 일반주제명
- Privacy
- 일반주제명
- Communication
- 일반주제명
- Neural networks
- 일반주제명
- Parameter estimation
- 기타저자
- Stanford University.
- 기본자료저록
- Dissertations Abstracts International. 86-05B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017164856
■00520250211153055
■006m o d
■007cr#unu||||||||
■020 ▼a9798346390039
■035 ▼a(MiAaPQ)AAI31643398
■035 ▼a(MiAaPQ)Stanfordwv299qm1978
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a300
■1001 ▼aIsik, Berivan.
■24510▼aStatistical Methods for Efficient and Trustworthy Machine Learning
■260 ▼a[Sl]▼bStanford University▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a268 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-05, Section: B.
■500 ▼aAdvisor: Weissman, Tsachy.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2024.
■520 ▼aThis thesis addresses key challenges in machine learning, with a focus on efficiency and trustworthiness through statistical tools. It explores emerging topics in the era of large models and big data, including model compression, federated learning, and data privacy.New methods for model and training data compression are introduced, leveraging rate-distortion theory and novel coding techniques designed for the specific demands of machine learning tasks. Additionally, new federated learning frameworks are proposed to reduce communication costs while maintaining or even enhancing accuracy and inference efficiency. These frameworks employ strategies such as model update sparsification, exploiting overparameterization in modern models, and utilizing side information to achieve substantial bitrate reductions without sacrificing performance.The research advances the understanding of compression-privacy-utility tradeoffs in training data processing and distributed mean estimation, offering new optimal algorithms that effectively balance these tradeoffs. These contributions lay a robust theoretical foundation for enhancing data privacy while reducing communication and storage costs, all without compromising utility.Overall, this work advances machine learning systems by proposing solutions that address the unique conditions, limitations, and flexibilities of these models, ultimately enhancing efficiency and trustworthiness across multiple domains.
■590 ▼aSchool code: 0212.
■650 4▼aPrivacy
■650 4▼aCommunication
■650 4▼aNeural networks
■650 4▼aParameter estimation
■690 ▼a0459
■690 ▼a0800
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g86-05B.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17164856▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
