본문

서브메뉴

On Safe Sequential Decision-Making in Adversarial Environments
On Safe Sequential Decision-Making in Adversarial Environments
On Safe Sequential Decision-Making in Adversarial Environments

상세정보

자료유형  
 학위논문 서양
최종처리일시  
20260202103537
ISBN  
9798288866173
DDC  
310
저자명  
Lekeufack Sopze, Jordan.
서명/저자  
On Safe Sequential Decision-Making in Adversarial Environments
발행사항  
[Sl] : University of California, Berkeley, 2025
발행사항  
Ann Arbor : ProQuest Dissertations & Theses, 2025
형태사항  
81 p
주기사항  
Source: Dissertations Abstracts International, Volume: 87-01, Section: B.
주기사항  
Advisor: Jordan, Michael I.
학위논문주기  
Thesis (Ph.D.)--University of California, Berkeley, 2025.
초록/해제  
요약Decision-making in adversarial environments is a critical challenge across various domains where agents must navigate uncertainty, competition, and potentially hostile conditions.In such settings, the ability to predict future outcomes plays a pivotal role in guiding decisions. However, predictions are rarely perfect, especially in adversarial environments where opposing forces might actively seek to disrupt or exploit weaknesses in an agent's strategy. The central challenge lies in ensuring safety---defined as avoiding catastrophic outcomes---despite the inherent imperfections in predictive models. This thesis addresses this dual objective of leveraging predictions to make informed decisions while simultaneously guaranteeing safety, even in the most adverse scenarios. The potential applications of safe sequential decision-making in adversarial environments are vast and span multiple domains. For instance, in robotics, autonomous agents operating in dynamic environments must avoid collisions or irreversible errors, even when interacting with other potentially hostile agents. Similarly, in finance, automated trading algorithms must operate in markets that are impacted by unpredictable events and the actions of other agents. This thesis contributes to these fields by developing algorithms for sequential decision-making providing theoretical guarantees for computationally efficient and practical algorithms.Chapter 1 will present some background and concepts that are used and referred to in the thesis. In particular, we introduce conformal prediction and online convex optimization.Chapter 2 introduces a new framework called Conformal Decision Theory.Drawing inspiration from online conformal prediction techniques, our approach offers a remarkably simple but powerful mechanism for guaranteeing long-term safety with minimal environmental constraints, contingent only on the existence of a safe decision.Chapter 3 provides an alternative approach that leverages the pre-existing online convex optimization framework with adversarial constraints, and shows how the existing guarantees can be significantly improved when using predictions. Importantly, even when the predictions are inaccurate, our approach still maintains guarantees that are on par with existing work.
일반주제명  
Statistics
일반주제명  
Computer science
키워드  
Conformal prediction
키워드  
Decision-making
키워드  
Online convex optimization
키워드  
Online learning
기타저자  
University of California, Berkeley Statistics
기본자료저록  
Dissertations Abstracts International. 87-01B.
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008260126s2025        us                              c    eng  d
■001000017357620
■00520260202103537
■006m          o    d                
■007cr#unu||||||||
■020    ▼a9798288866173
■035    ▼a(MiAaPQ)AAI32040601
■040    ▼aMiAaPQ▼cMiAaPQ
■0820  ▼a310
■1001  ▼aLekeufack  Sopze,  Jordan.
■24510▼aOn  Safe  Sequential  Decision-Making  in  Adversarial  Environments
■260    ▼a[Sl]▼bUniversity  of  California,  Berkeley▼c2025
■260  1▼aAnn  Arbor▼bProQuest  Dissertations  &  Theses▼c2025
■300    ▼a81  p
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  87-01,  Section:  B.
■500    ▼aAdvisor:  Jordan,  Michael  I.
■5021  ▼aThesis  (Ph.D.)--University  of  California,  Berkeley,  2025.
■520    ▼aDecision-making  in  adversarial  environments  is  a  critical  challenge  across  various  domains  where  agents  must  navigate  uncertainty,  competition,  and  potentially  hostile  conditions.In  such  settings,  the  ability  to  predict  future  outcomes  plays  a  pivotal  role  in  guiding  decisions.  However,  predictions  are  rarely  perfect,  especially  in  adversarial  environments  where  opposing  forces  might  actively  seek  to  disrupt  or  exploit  weaknesses  in  an  agent's  strategy.  The  central  challenge  lies  in  ensuring  safety---defined  as  avoiding  catastrophic  outcomes---despite  the  inherent  imperfections  in  predictive  models.  This  thesis  addresses  this  dual  objective  of  leveraging  predictions  to  make  informed  decisions  while  simultaneously  guaranteeing  safety,  even  in  the  most  adverse  scenarios.  The  potential  applications  of  safe  sequential  decision-making  in  adversarial  environments  are  vast  and  span  multiple  domains.  For  instance,  in  robotics,  autonomous  agents  operating  in  dynamic  environments  must  avoid  collisions  or  irreversible  errors,  even  when  interacting  with  other  potentially  hostile  agents.  Similarly,  in  finance,  automated  trading  algorithms  must  operate  in  markets  that  are  impacted  by  unpredictable  events  and  the  actions  of  other  agents.  This  thesis  contributes  to  these  fields  by  developing  algorithms  for  sequential  decision-making  providing  theoretical  guarantees  for  computationally  efficient  and  practical  algorithms.Chapter  1  will  present  some  background  and  concepts  that  are  used  and  referred  to  in  the  thesis.  In  particular,  we  introduce  conformal  prediction  and  online  convex  optimization.Chapter  2  introduces  a  new  framework  called  Conformal  Decision  Theory.Drawing  inspiration  from  online  conformal  prediction  techniques,  our  approach  offers  a  remarkably  simple  but  powerful  mechanism  for  guaranteeing  long-term  safety  with  minimal  environmental  constraints,  contingent  only  on  the  existence  of  a  safe  decision.Chapter  3  provides  an  alternative  approach  that  leverages  the  pre-existing  online  convex  optimization  framework  with  adversarial  constraints,  and  shows  how  the  existing  guarantees  can  be  significantly  improved  when  using  predictions.  Importantly,  even  when  the  predictions  are  inaccurate,  our  approach  still  maintains  guarantees  that  are  on  par  with  existing  work.
■590    ▼aSchool  code:  0028.
■650  4▼aStatistics
■650  4▼aComputer  science
■653    ▼aConformal  prediction
■653    ▼aDecision-making
■653    ▼aOnline  convex  optimization
■653    ▼aOnline  learning
■690    ▼a0463
■690    ▼a0796
■690    ▼a0984
■71020▼aUniversity  of  California,  Berkeley▼bStatistics.
■7730  ▼tDissertations  Abstracts  International▼g87-01B.
■790    ▼a0028
■791    ▼aPh.D.
■792    ▼a2025
■793    ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17357620▼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
    TF18296 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.