본문

서브메뉴

Trustworthy Machine Learning: Privacy-Preserving and Adversarially Robust Optimization Algorithms
Trustworthy Machine Learning: Privacy-Preserving and Adversarially Robust Optimization Alg...
Trustworthy Machine Learning: Privacy-Preserving and Adversarially Robust Optimization Algorithms

상세정보

자료유형  
 학위논문 서양
최종처리일시  
20260202105130
ISBN  
9798291553381
DDC  
004
저자명  
Gao, Changyu.
서명/저자  
Trustworthy Machine Learning: Privacy-Preserving and Adversarially Robust Optimization Algorithms
발행사항  
[Sl] : The University of Wisconsin - Madison, 2025
발행사항  
Ann Arbor : ProQuest Dissertations & Theses, 2025
형태사항  
232 p
주기사항  
Source: Dissertations Abstracts International, Volume: 87-02, Section: B.
주기사항  
Advisor: Wright, Stephen.
학위논문주기  
Thesis (Ph.D.)--The University of Wisconsin - Madison, 2025.
초록/해제  
요약This thesis investigates the development of trustworthy machine learning systems by focusing on two critical areas: privacy and robustness. We introduce and analyze novel optimization algorithms designed to perform effectively under the rigorous constraints of differential privacy and in the presence of adversarial data perturbations. The central aim is to design algorithms that are not only theoretically sound but also practically efficient, addressing fundamental trade-offs between privacy, robustness, and computational performance in modern machine learning.Key contributions of this work include the following: the development of a practical and efficient framework for differentially private nonconvex optimization (Chapter 2) that finds approximate second-order solutions. In the distributed setting, we present novel algorithms for federated learning (Chapter 3) that achieve optimal excess risk bounds with superior communication efficiency for heterogeneous data, resolving a previously open problem. To address adversarial data corruption, in Chapter 4, we develop a novel algorithm for robust stochastic convex optimization under the strong contamination model that achieves minimax-optimal excess risk. The work also provides a broader perspective on the landscape of private learning by summarizing key sample complexity results and charting future research directions (Chapter 5). We provide partial results to these open problems. Notably we present the first algorithm achieving both differential privacy and adversarial robustness simultaneously for stochastic convex optimization.In summary, this dissertation advances the state-of-the-art by providing new algorithms with strong theoretical guarantees and practical advantages for private and robust machine learning. The findings presented offer a deeper understanding of the fundamental trade-offs between privacy, robustness, and computational efficiency, paving the way for more reliable and secure machine learning systems.
일반주제명  
Computer science
일반주제명  
Computer engineering
키워드  
Adversarial robustness
키워드  
Differential privacy
키워드  
Mathematical optimization
키워드  
Nonsmooth loss
키워드  
Trustworthy machine learning
기타저자  
The University of Wisconsin - Madison Industrial Engineering
기본자료저록  
Dissertations Abstracts International. 87-02B.
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008260126s2025        us                              c    eng  d
■001000017359513
■00520260202105130
■006m          o    d                
■007cr#unu||||||||
■020    ▼a9798291553381
■035    ▼a(MiAaPQ)AAI32239182
■040    ▼aMiAaPQ▼cMiAaPQ
■0820  ▼a004
■1001  ▼aGao,  Changyu.
■24510▼aTrustworthy  Machine  Learning:  Privacy-Preserving  and  Adversarially  Robust  Optimization  Algorithms
■260    ▼a[Sl]▼bThe  University  of  Wisconsin  -  Madison▼c2025
■260  1▼aAnn  Arbor▼bProQuest  Dissertations  &  Theses▼c2025
■300    ▼a232  p
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  87-02,  Section:  B.
■500    ▼aAdvisor:  Wright,  Stephen.
■5021  ▼aThesis  (Ph.D.)--The  University  of  Wisconsin  -  Madison,  2025.
■520    ▼aThis  thesis  investigates  the  development  of  trustworthy  machine  learning  systems  by  focusing  on  two  critical  areas:  privacy  and  robustness.  We  introduce  and  analyze  novel  optimization  algorithms  designed  to  perform  effectively  under  the  rigorous  constraints  of  differential  privacy  and  in  the  presence  of  adversarial  data  perturbations.  The  central  aim  is  to  design  algorithms  that  are  not  only  theoretically  sound  but  also  practically  efficient,  addressing  fundamental  trade-offs  between  privacy,  robustness,  and  computational  performance  in  modern  machine  learning.Key  contributions  of  this  work  include  the  following:  the  development  of  a  practical  and  efficient  framework  for  differentially  private  nonconvex  optimization  (Chapter  2)  that  finds  approximate  second-order  solutions.  In  the  distributed  setting,  we  present  novel  algorithms  for  federated  learning  (Chapter  3)  that  achieve  optimal  excess  risk  bounds  with  superior  communication  efficiency  for  heterogeneous  data,  resolving  a  previously  open  problem.  To  address  adversarial  data  corruption,  in  Chapter  4,  we  develop  a  novel  algorithm  for  robust  stochastic  convex  optimization  under  the  strong  contamination  model  that  achieves  minimax-optimal  excess  risk.  The  work  also  provides  a  broader  perspective  on  the  landscape  of  private  learning  by  summarizing  key  sample  complexity  results  and  charting  future  research  directions  (Chapter  5).  We  provide  partial  results  to  these  open  problems.  Notably  we  present  the  first  algorithm  achieving  both  differential  privacy  and  adversarial  robustness  simultaneously  for  stochastic  convex  optimization.In  summary,  this  dissertation  advances  the  state-of-the-art  by  providing  new  algorithms  with  strong  theoretical  guarantees  and  practical  advantages  for  private  and  robust  machine  learning.  The  findings  presented  offer  a  deeper  understanding  of  the  fundamental  trade-offs  between  privacy,  robustness,  and  computational  efficiency,  paving  the  way  for  more  reliable  and  secure  machine  learning  systems.
■590    ▼aSchool  code:  0262.
■650  4▼aComputer  science
■650  4▼aComputer  engineering
■653    ▼aAdversarial  robustness
■653    ▼aDifferential  privacy
■653    ▼aMathematical  optimization
■653    ▼aNonsmooth  loss
■653    ▼aTrustworthy  machine  learning
■690    ▼a0796
■690    ▼a0984
■690    ▼a0464
■71020▼aThe  University  of  Wisconsin  -  Madison▼bIndustrial  Engineering.
■7730  ▼tDissertations  Abstracts  International▼g87-02B.
■790    ▼a0262
■791    ▼aPh.D.
■792    ▼a2025
■793    ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17359513▼nKERIS▼z이  자료의  원문은  한국교육학술정보원에서  제공합니다.

미리보기

내보내기

chatGPT토론

Ai 추천 관련 도서


    신착도서 더보기
    최근 3년간 통계입니다.

    소장정보

    • 예약
    • 소재불명신고
    • 나의폴더
    • 우선정리요청
    • 비도서대출신청
    • 야간 도서대출신청
    소장자료
    등록번호 청구기호 소장처 대출가능여부 대출정보
    TF16683 전자도서 대출가능 마이폴더 부재도서신고 비도서대출신청 야간 도서대출신청

    * 대출중인 자료에 한하여 예약이 가능합니다. 예약을 원하시면 예약버튼을 클릭하십시오.

    해당 도서를 다른 이용자가 함께 대출한 도서

    관련 인기도서

    로그인 후 이용 가능합니다.