본문

서브메뉴

Deep Generative Models for Decision-Making and Control- [electronic resource]
Deep Generative Models for Decision-Making and Control - [electronic resource]
Deep Generative Models for Decision-Making and Control- [electronic resource]

상세정보

자료유형  
 학위논문파일 국외
최종처리일시  
20240214100500
ISBN  
9798380380911
DDC  
629.8
저자명  
Janner, Michael.
서명/저자  
Deep Generative Models for Decision-Making and Control - [electronic resource]
발행사항  
[S.l.]: : University of California, Berkeley., 2023
발행사항  
Ann Arbor : : ProQuest Dissertations & Theses,, 2023
형태사항  
1 online resource(98 p.)
주기사항  
Source: Dissertations Abstracts International, Volume: 85-03, Section: B.
주기사항  
Advisor: Levine, Sergey.
학위논문주기  
Thesis (Ph.D.)--University of California, Berkeley, 2023.
사용제한주기  
This item must not be sold to any third party vendors.
초록/해제  
요약Deep model-based reinforcement learning methods offer a conceptually simple approach to the decision-making and control problem: use learning for the purpose of estimating an approximate dynamics model, and offload the rest of the work to classical trajectory optimization. However, this combination has a number of empirical shortcomings, limiting the usefulness of model-based methods in practice. The dual purpose of this thesis is to study the reasons for these shortcomings and to propose solutions for the uncovered problems. We begin by generalizing the dynamics model itself, replacing the standard single-step formulation with a model that predicts over probabilistic latent horizons. The resulting model, trained with a generative reinterpretation of temporal difference learning, leads to infinite-horizon variants of the procedures central to model-based control, including the model rollout and model-based value estimation.Next, we show that poor predictive accuracy of commonly-used deep dynamics models is a major bottleneck to effective planning, and describe how to use high-capacity sequence models to overcome this limitation. Framing reinforcement learning as sequence modeling simplifies a range of design decisions, allowing us to dispense with many of the components normally integral to reinforcement learning algorithms. However, despite their predictive accuracy, such sequence models are limited by the search algorithms in which they are embedded. As such, we demonstrate how to fold the entire trajectory optimization pipeline into the generative model itself, such that sampling from the model and planning with it become nearly identical. The culmination of this endeavor is a method that improves its planning capabilities, and not just its predictive accuracy, with more data and experience. Along the way, we highlight how inference techniques from the contemporary generative modeling toolbox, including beam search, classifier-guided sampling, and image inpainting, can be reinterpreted as viable planning strategies for reinforcement learning problems.
일반주제명  
Robotics.
일반주제명  
Computer science.
키워드  
Reinforcement learning
키워드  
Dynamics model
키워드  
Single-step formulation
기타저자  
University of California, Berkeley Computer Science
기본자료저록  
Dissertations Abstracts International. 85-03B.
기본자료저록  
Dissertation Abstract International
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008240612s2023      us  |||||||||||||||c||eng  d
■001000016932454
■00520240214100500
■006m          o    d                
■007cr#unu||||||||
■020    ▼a9798380380911
■035    ▼a(MiAaPQ)AAI30492754
■040    ▼aMiAaPQ▼cMiAaPQ
■0820  ▼a629.8
■1001  ▼aJanner,  Michael.
■24510▼aDeep  Generative  Models  for  Decision-Making  and  Control▼h[electronic  resource]
■260    ▼a[S.l.]:▼bUniversity  of  California,  Berkeley.  ▼c2023
■260  1▼aAnn  Arbor  :▼bProQuest  Dissertations  &  Theses,  ▼c2023
■300    ▼a1  online  resource(98  p.)
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  85-03,  Section:  B.
■500    ▼aAdvisor:  Levine,  Sergey.
■5021  ▼aThesis  (Ph.D.)--University  of  California,  Berkeley,  2023.
■506    ▼aThis  item  must  not  be  sold  to  any  third  party  vendors.
■520    ▼aDeep  model-based  reinforcement  learning  methods  offer  a  conceptually  simple  approach  to  the  decision-making  and  control  problem:  use  learning  for  the  purpose  of  estimating  an  approximate  dynamics  model,  and  offload  the  rest  of  the  work  to  classical  trajectory  optimization.  However,  this  combination  has  a  number  of  empirical  shortcomings,  limiting  the  usefulness  of  model-based  methods  in  practice.  The  dual  purpose  of  this  thesis  is  to  study  the  reasons  for  these  shortcomings  and  to  propose  solutions  for  the  uncovered  problems.  We  begin  by  generalizing  the  dynamics  model  itself,  replacing  the  standard  single-step  formulation  with  a  model  that  predicts  over  probabilistic  latent  horizons.  The  resulting  model,  trained  with  a  generative  reinterpretation  of  temporal  difference  learning,  leads  to  infinite-horizon  variants  of  the  procedures  central  to  model-based  control,  including  the  model  rollout  and  model-based  value  estimation.Next,  we  show  that  poor  predictive  accuracy  of  commonly-used  deep  dynamics  models  is  a  major  bottleneck  to  effective  planning,  and  describe  how  to  use  high-capacity  sequence  models  to  overcome  this  limitation.  Framing  reinforcement  learning  as  sequence  modeling  simplifies  a  range  of  design  decisions,  allowing  us  to  dispense  with  many  of  the  components  normally  integral  to  reinforcement  learning  algorithms.  However,  despite  their  predictive  accuracy,  such  sequence  models  are  limited  by  the  search  algorithms  in  which  they  are  embedded.  As  such,  we  demonstrate  how  to  fold  the  entire  trajectory  optimization  pipeline  into  the  generative  model  itself,  such  that  sampling  from  the  model  and  planning  with  it  become  nearly  identical.  The  culmination  of  this  endeavor  is  a  method  that  improves  its  planning  capabilities,  and  not  just  its  predictive  accuracy,  with  more  data  and  experience.  Along  the  way,  we  highlight  how  inference  techniques  from  the  contemporary  generative  modeling  toolbox,  including  beam  search,  classifier-guided  sampling,  and  image  inpainting,  can  be  reinterpreted  as  viable  planning  strategies  for  reinforcement  learning  problems.
■590    ▼aSchool  code:  0028.
■650  4▼aRobotics.
■650  4▼aComputer  science.
■653    ▼aReinforcement  learning
■653    ▼aDynamics  model
■653    ▼aSingle-step  formulation
■690    ▼a0800
■690    ▼a0771
■690    ▼a0984
■71020▼aUniversity  of  California,  Berkeley▼bComputer  Science.
■7730  ▼tDissertations  Abstracts  International▼g85-03B.
■773    ▼tDissertation  Abstract  International
■790    ▼a0028
■791    ▼aPh.D.
■792    ▼a2023
■793    ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T16932454▼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
    TF09411 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.