본문

서브메뉴

Leveraging Compositional Structure for Reinforcement Learning and Sequential Decision Making Problems
Leveraging Compositional Structure for Reinforcement Learning and Sequential Decision Maki...
Leveraging Compositional Structure for Reinforcement Learning and Sequential Decision Making Problems

Detailed Information

자료유형  
 학위논문 서양
최종처리일시  
20260202103648
ISBN  
9798314875445
DDC  
004
저자명  
Liu, Anthony Z.
서명/저자  
Leveraging Compositional Structure for Reinforcement Learning and Sequential Decision Making Problems
발행사항  
[Sl] : University of Michigan, 2025
발행사항  
Ann Arbor : ProQuest Dissertations & Theses, 2025
형태사항  
149 p
주기사항  
Source: Dissertations Abstracts International, Volume: 86-11, Section: A.
주기사항  
Advisor: Lee, Honglak.
학위논문주기  
Thesis (Ph.D.)--University of Michigan, 2025.
초록/해제  
요약Deep Learning approaches have made tremendous progress toward solving reinforcement learning and sequential decision-making problems. However, current approaches still struggle with long-horizon tasks that require strong generalization. These are tasks that an agent must solve using many actions and may have situations where the agent must generalize its actions from prior experience. A dominant approach is to solve these tasks in a hierarchical manner: a high-level agent decomposes a task into multiple "subtasks" to be individually solved by a low-level agent, which specializes in solving these subtasks. The effectiveness of this approach is enhanced by identifying and utilizing the inherent compositional structures of tasks, which enable more efficient learning and broader generalization.This dissertation introduces novel methodologies that build on task compositionality to address these challenges. Key contributions include: (1) Higher-Order Skill Learning: A hierarchical reinforcement learning framework is proposed, enabling low-level policies to optimize for sequences of subtasks rather than individual ones, resulting in improved efficiency and performance. (2) Parameterized Task Structures: Introducing parameterized subtask graphs to model tasks with compositional structures, enhancing both the efficiency of task inference and generalization to unseen entities. In addition, contributions that show compositional structure can be used through language and large language models (LLMs): (3) Integrating Multimodal Observations in Language Models: Demonstrating that visual observations can be embedded as input tokens for large language models (LLMs), achieving state-of-the-art performance in visually grounded planning tasks. (4) Skill Abstractions for LLM Planning: Highlighting the benefits of providing LLMs with structured descriptions of subtasks, or skills, to improve planning and reasoning capabilities. (5) Code-Augmented Planning: Proposing a method where LLMs use control flow constructs to generate and execute code for solving complex planning tasks, significantly improving task performance.Collectively, these approaches showcase how leveraging task compositionality through hierarchical structures and language from LLMs can improve reinforcement learning and sequential decision-making frameworks.
일반주제명  
Computer science
일반주제명  
Computer engineering
일반주제명  
Information science
키워드  
Hierarchical reinforcement learning
키워드  
Task decomposition
키워드  
Large language models
키워드  
Task generalization
키워드  
Decision-making problems
기타저자  
University of Michigan Computer Science & Engineering
기본자료저록  
Dissertations Abstracts International. 86-11A.
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008260126s2025        us                              c    eng  d
■001000017358126
■00520260202103648
■006m          o    d                
■007cr#unu||||||||
■020    ▼a9798314875445
■035    ▼a(MiAaPQ)AAI32092652
■035    ▼a(MiAaPQ)umichrackham005989
■040    ▼aMiAaPQ▼cMiAaPQ
■0820  ▼a004
■1001  ▼aLiu,  Anthony  Z.
■24510▼aLeveraging  Compositional  Structure  for  Reinforcement  Learning  and  Sequential  Decision  Making  Problems
■260    ▼a[Sl]▼bUniversity  of  Michigan▼c2025
■260  1▼aAnn  Arbor▼bProQuest  Dissertations  &  Theses▼c2025
■300    ▼a149  p
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  86-11,  Section:  A.
■500    ▼aAdvisor:  Lee,  Honglak.
■5021  ▼aThesis  (Ph.D.)--University  of  Michigan,  2025.
■520    ▼aDeep  Learning  approaches  have  made  tremendous  progress  toward  solving  reinforcement  learning  and  sequential  decision-making  problems.  However,  current  approaches  still  struggle  with  long-horizon  tasks  that  require  strong  generalization.  These  are  tasks  that  an  agent  must  solve  using  many  actions  and  may  have  situations  where  the  agent  must  generalize  its  actions  from  prior  experience.  A  dominant  approach  is  to  solve  these  tasks  in  a  hierarchical  manner:  a  high-level  agent  decomposes  a  task  into  multiple  "subtasks"  to  be  individually  solved  by  a  low-level  agent,  which  specializes  in  solving  these  subtasks.  The  effectiveness  of  this  approach  is  enhanced  by  identifying  and  utilizing  the  inherent  compositional  structures  of  tasks,  which  enable  more  efficient  learning  and  broader  generalization.This  dissertation  introduces  novel  methodologies  that  build  on  task  compositionality  to  address  these  challenges.  Key  contributions  include:  (1)  Higher-Order  Skill  Learning:  A  hierarchical  reinforcement  learning  framework  is  proposed,  enabling  low-level  policies  to  optimize  for  sequences  of  subtasks  rather  than  individual  ones,  resulting  in  improved  efficiency  and  performance.  (2)  Parameterized  Task  Structures:  Introducing  parameterized  subtask  graphs  to  model  tasks  with  compositional  structures,  enhancing  both  the  efficiency  of  task  inference  and  generalization  to  unseen  entities.  In  addition,  contributions  that  show  compositional  structure  can  be  used  through  language  and  large  language  models  (LLMs):  (3)  Integrating  Multimodal  Observations  in  Language  Models:  Demonstrating  that  visual  observations  can  be  embedded  as  input  tokens  for  large  language  models  (LLMs),  achieving  state-of-the-art  performance  in  visually  grounded  planning  tasks.  (4)  Skill  Abstractions  for  LLM  Planning:  Highlighting  the  benefits  of  providing  LLMs  with  structured  descriptions  of  subtasks,  or  skills,  to  improve  planning  and  reasoning  capabilities.  (5)  Code-Augmented  Planning:  Proposing  a  method  where  LLMs  use  control  flow  constructs  to  generate  and  execute  code  for  solving  complex  planning  tasks,  significantly  improving  task  performance.Collectively,  these  approaches  showcase  how  leveraging  task  compositionality  through  hierarchical  structures  and  language  from  LLMs  can  improve  reinforcement  learning  and  sequential  decision-making  frameworks.
■590    ▼aSchool  code:  0127.
■650  4▼aComputer  science
■650  4▼aComputer  engineering
■650  4▼aInformation  science
■653    ▼aHierarchical  reinforcement  learning
■653    ▼aTask  decomposition
■653    ▼aLarge  language  models
■653    ▼aTask  generalization
■653    ▼aDecision-making  problems
■690    ▼a0984
■690    ▼a0464
■690    ▼a0723
■71020▼aUniversity  of  Michigan▼bComputer  Science  &  Engineering.
■7730  ▼tDissertations  Abstracts  International▼g86-11A.
■790    ▼a0127
■791    ▼aPh.D.
■792    ▼a2025
■793    ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17358126▼nKERIS▼z이  자료의  원문은  한국교육학술정보원에서  제공합니다.

Preview

Export

ChatGPT Discussion

AI Recommended Related Books


    New Books MORE
    Statistics for the past 3 years. Go to brief

    Info Détail de la recherche.

    • Réservation
    • n'existe pas
    • My Folder
    • Demande Première utilisation
    • Non-Book Loan Application
    • Nighttime Book Loan Application
    Matériel
    Reg No. Call No. emplacement Status Lend Info
    TF16006 전자도서 대출가능 My Folder 부재도서신고 비도서대출신청 야간 도서대출신청

    * Les réservations sont disponibles dans le livre d'emprunt. Pour faire des réservations, S'il vous plaît cliquer sur le bouton de réservation

    Books borrowed together with this book

    Related Popular Books

    Available after logging in.