본문

서브메뉴

Privacy and Efficiency in Personalized Decision-Making and Recommendation- [electronic resource]
Privacy and Efficiency in Personalized Decision-Making and Recommendation - [electronic re...
Privacy and Efficiency in Personalized Decision-Making and Recommendation- [electronic resource]

상세정보

자료유형  
 학위논문파일 국외
최종처리일시  
20240214101625
ISBN  
9798380469920
DDC  
658
저자명  
Carranza, Aldo Gael.
서명/저자  
Privacy and Efficiency in Personalized Decision-Making and Recommendation - [electronic resource]
발행사항  
[S.l.]: : Stanford University., 2023
발행사항  
Ann Arbor : : ProQuest Dissertations & Theses,, 2023
형태사항  
1 online resource(218 p.)
주기사항  
Source: Dissertations Abstracts International, Volume: 85-04, Section: B.
주기사항  
Advisor: Athey, Susan.
학위논문주기  
Thesis (Ph.D.)--Stanford University, 2023.
사용제한주기  
This item must not be sold to any third party vendors.
초록/해제  
요약In the current digital era, marked by the ubiquity of individual-level data and sophisticated artificial intelligence systems highly capable of exploiting data heterogeneity, data-driven personalized decision-making and recommendation systems have become prevalent in providing customized services and experiences to individuals. These bespoke systems, however, continue to present considerable challenges regarding their deployment in adaptive, heterogeneous, and privacy-sensitive settings. This dissertation presents three research projects that delve into some of these critical issues, offering insights and novel solutions aimed at enhancing the privacy and efficiency of personalized decision-making and recommendation systems.Chapter 1 of this dissertation investigates the challenges of model learning in contextual bandits for adaptive decision-making and presents a method to improve data efficiency and robustness to model misspecification in this online setting. Contextual bandit algorithms often estimate reward models to inform decision-making. However, true rewards can contain action-independent redundancies that are not relevant for decision-making. We show it is more data-efficient to estimate any function that explains the reward differences between actions, that is, the treatment effects. Motivated by this observation, building on recent work on oracle-based bandit algorithms, we provide a universal reduction of contextual bandits to general-purpose heterogeneous treatment effect estimation, and we design a simple and computationally efficient algorithm based on this reduction. Our theoretical and experimental results demonstrate that heterogeneous treatment effect estimation in contextual bandits offers practical advantages over reward estimation, including more efficient model estimation and greater robustness to model misspecification.In Chapter 2, we consider heterogeneous data adaptation and privacy in decision-making informed by historical observational data. We consider the problem of learning personalized decision policies on observational bandit feedback data from heterogeneous data sources. Moreover, we examine this problem in the federated setting where a central server aims to learn a policy on the data distributed across the heterogeneous sources without exchanging their raw data due to privacy concerns. We present a federated policy learning algorithm based on aggregation of local policies trained with doubly robust offline policy evaluation and learning strategies. We provide a novel regret analysis for our approach that establishes a finite-sample upper bound on a notion of global regret across a distribution of clients. In addition, for any individual client, we establish a corresponding local regret upper bound characterized by the presence of distribution shift relative to all other clients. We support our theoretical findings with experimental results. Our analysis and experiments provide insights into the value of heterogeneous client participation in federation for policy learning in heterogeneous settings.Lastly, in Chapter 3, we pivot slightly from the focus of the first two chapters on decision-making systems with online and offline policy learning methods to investigating data privacy in recommender systems. We propose a novel approach for developing privacy-preserving large-scale recommender systems using differentially private (DP) large language models (LLMs) which overcomes certain challenges and limitations in DP training these complex systems. This method is particularly well suited for the emerging area of LLM-based recommender systems, but can be readily employed for any recommender systems that process representations of natural language inputs. Our approach involves using DP training methods to fine-tune a publicly pre-trained LLM on a query generation task.
일반주제명  
Decision making.
일반주제명  
Computer science.
일반주제명  
Recommender systems.
기타저자  
Stanford University.
기본자료저록  
Dissertations Abstracts International. 85-04B.
기본자료저록  
Dissertation Abstract International
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008240612s2023      us  |||||||||||||||c||eng  d
■001000016934554
■00520240214101625
■006m          o    d                
■007cr#unu||||||||
■020    ▼a9798380469920
■035    ▼a(MiAaPQ)AAI30615162
■035    ▼a(MiAaPQ)STANFORDky391xq2715
■040    ▼aMiAaPQ▼cMiAaPQ
■0820  ▼a658
■1001  ▼aCarranza,  Aldo  Gael.
■24510▼aPrivacy  and  Efficiency  in  Personalized  Decision-Making  and  Recommendation▼h[electronic  resource]
■260    ▼a[S.l.]:▼bStanford  University.  ▼c2023
■260  1▼aAnn  Arbor  :▼bProQuest  Dissertations  &  Theses,  ▼c2023
■300    ▼a1  online  resource(218  p.)
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  85-04,  Section:  B.
■500    ▼aAdvisor:  Athey,  Susan.
■5021  ▼aThesis  (Ph.D.)--Stanford  University,  2023.
■506    ▼aThis  item  must  not  be  sold  to  any  third  party  vendors.
■520    ▼aIn  the  current  digital  era,  marked  by  the  ubiquity  of  individual-level  data  and  sophisticated  artificial  intelligence  systems  highly  capable  of  exploiting  data  heterogeneity,  data-driven  personalized  decision-making  and  recommendation  systems  have  become  prevalent  in  providing  customized  services  and  experiences  to  individuals.  These  bespoke  systems,  however,  continue  to  present  considerable  challenges  regarding  their  deployment  in  adaptive,  heterogeneous,  and  privacy-sensitive  settings.  This  dissertation  presents  three  research  projects  that  delve  into  some  of  these  critical  issues,  offering  insights  and  novel  solutions  aimed  at  enhancing  the  privacy  and  efficiency  of  personalized  decision-making  and  recommendation  systems.Chapter  1  of  this  dissertation  investigates  the  challenges  of  model  learning  in  contextual  bandits  for  adaptive  decision-making  and  presents  a  method  to  improve  data  efficiency  and  robustness  to  model  misspecification  in  this  online  setting.  Contextual  bandit  algorithms  often  estimate  reward  models  to  inform  decision-making.  However,  true  rewards  can  contain  action-independent  redundancies  that  are  not  relevant  for  decision-making.  We  show  it  is  more  data-efficient  to  estimate  any  function  that  explains  the  reward  differences  between  actions,  that  is,  the  treatment  effects.  Motivated  by  this  observation,  building  on  recent  work  on  oracle-based  bandit  algorithms,  we  provide  a  universal  reduction  of  contextual  bandits  to  general-purpose  heterogeneous  treatment  effect  estimation,  and  we  design  a  simple  and  computationally  efficient  algorithm  based  on  this  reduction.  Our  theoretical  and  experimental  results  demonstrate  that  heterogeneous  treatment  effect  estimation  in  contextual  bandits  offers  practical  advantages  over  reward  estimation,  including  more  efficient  model  estimation  and  greater  robustness  to  model  misspecification.In  Chapter  2,  we  consider  heterogeneous  data  adaptation  and  privacy  in  decision-making  informed  by  historical  observational  data.  We  consider  the  problem  of  learning  personalized  decision  policies  on  observational  bandit  feedback  data  from  heterogeneous  data  sources.  Moreover,  we  examine  this  problem  in  the  federated  setting  where  a  central  server  aims  to  learn  a  policy  on  the  data  distributed  across  the  heterogeneous  sources  without  exchanging  their  raw  data  due  to  privacy  concerns.  We  present  a  federated  policy  learning  algorithm  based  on  aggregation  of  local  policies  trained  with  doubly  robust  offline  policy  evaluation  and  learning  strategies.  We  provide  a  novel  regret  analysis  for  our  approach  that  establishes  a  finite-sample  upper  bound  on  a  notion  of  global  regret  across  a  distribution  of  clients.  In  addition,  for  any  individual  client,  we  establish  a  corresponding  local  regret  upper  bound  characterized  by  the  presence  of  distribution  shift  relative  to  all  other  clients.  We  support  our  theoretical  findings  with  experimental  results.  Our  analysis  and  experiments  provide  insights  into  the  value  of  heterogeneous  client  participation  in  federation  for  policy  learning  in  heterogeneous  settings.Lastly,  in  Chapter  3,  we  pivot  slightly  from  the  focus  of  the  first  two  chapters  on  decision-making  systems  with  online  and  offline  policy  learning  methods  to  investigating  data  privacy  in  recommender  systems.  We  propose  a  novel  approach  for  developing  privacy-preserving  large-scale  recommender  systems  using  differentially  private  (DP)  large  language  models  (LLMs)  which  overcomes  certain  challenges  and  limitations  in  DP  training  these  complex  systems.  This  method  is  particularly  well  suited  for  the  emerging  area  of  LLM-based  recommender  systems,  but  can  be  readily  employed  for  any  recommender  systems  that  process  representations  of  natural  language  inputs.  Our  approach  involves  using  DP  training  methods  to  fine-tune  a  publicly  pre-trained  LLM  on  a  query  generation  task.
■590    ▼aSchool  code:  0212.
■650  4▼aDecision  making.
■650  4▼aComputer  science.
■650  4▼aRecommender  systems.
■690    ▼a0984
■690    ▼a0800
■71020▼aStanford  University.
■7730  ▼tDissertations  Abstracts  International▼g85-04B.
■773    ▼tDissertation  Abstract  International
■790    ▼a0212
■791    ▼aPh.D.
■792    ▼a2023
■793    ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T16934554▼nKERIS▼z이  자료의  원문은  한국교육학술정보원에서  제공합니다.
■980    ▼a202402▼f2024

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
    TF08657 전자도서 마이폴더 부재도서신고 비도서대출신청

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

    해당 도서를 다른 이용자가 함께 대출한 도서

    관련 인기도서

    로그인 후 이용 가능합니다.