서브메뉴
검색
Leveraging Cross-Task Transfer in Sequential Decision Problems
Leveraging Cross-Task Transfer in Sequential Decision Problems
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20250211152018
- ISBN
- 9798383057247
- DDC
- 004
- 저자명
- Zentner, K. R.
- 서명/저자
- Leveraging Cross-Task Transfer in Sequential Decision Problems
- 발행사항
- [Sl] : University of Southern California, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 111 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 85-12, Section: B.
- 주기사항
- Advisor: Sukhatme, Gaurav S.
- 학위논문주기
- Thesis (Ph.D.)--University of Southern California, 2024.
- 초록/해제
- 요약The past few years have seen an explosion of interest in using machine learning to make robots capable of learning a diverse set of tasks. These robots use Reinforcement Learning to learn detailed sub-second interactions, but consequently require large amounts of data for each task. In this thesis we explore how Reinforcement Learning can be combined with Transfer Learning to re-use data across tasks. We begin by reviewing the start of Multi-Task and Meta RL and describe the motivations for using Transfer Learning. Then, we describe a basic framework for using Transfer Learning to efficiently learn multiple tasks, and show how it requires predicting how effectively transfer can be performed across tasks. Next, we present a simple rule, based in information theory, for predicting the effectiveness of Cross-Task Transfer. We discuss the theoretical implications of that rule, and show various quantitative evaluations of it. Then, we show two directions of work making use of our insights to perform efficient Transfer Reinforcement Learning. The first of these directions uses Cross-Task Co-Learning and Plan Conditioned Behavioral Cloning to share skill representations produced by a Large Language Model, and it able to learn many tasks from a single demonstration each in a simulated environment. The second of these directions uses Two-Phase KL Penalization to enforce a (potentially off-policy) trust region. These advances in Transfer RL may enable robots to be used in a wider range of applications, and may also inform applying Transfer RL outside of robotics.
- 일반주제명
- Computer science
- 일반주제명
- Robotics
- 일반주제명
- Information technology
- 키워드
- Machine learning
- 기타저자
- University of Southern California Computer Science
- 기본자료저록
- Dissertations Abstracts International. 85-12B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008250123s2024 us c eng d■001000017162492
■00520250211152018
■006m o d
■007cr#unu||||||||
■020 ▼a9798383057247
■035 ▼a(MiAaPQ)AAI31331965
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a004
■1001 ▼aZentner, K. R.
■24510▼aLeveraging Cross-Task Transfer in Sequential Decision Problems
■260 ▼a[Sl]▼bUniversity of Southern California▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a111 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 85-12, Section: B.
■500 ▼aAdvisor: Sukhatme, Gaurav S.
■5021 ▼aThesis (Ph.D.)--University of Southern California, 2024.
■520 ▼aThe past few years have seen an explosion of interest in using machine learning to make robots capable of learning a diverse set of tasks. These robots use Reinforcement Learning to learn detailed sub-second interactions, but consequently require large amounts of data for each task. In this thesis we explore how Reinforcement Learning can be combined with Transfer Learning to re-use data across tasks. We begin by reviewing the start of Multi-Task and Meta RL and describe the motivations for using Transfer Learning. Then, we describe a basic framework for using Transfer Learning to efficiently learn multiple tasks, and show how it requires predicting how effectively transfer can be performed across tasks. Next, we present a simple rule, based in information theory, for predicting the effectiveness of Cross-Task Transfer. We discuss the theoretical implications of that rule, and show various quantitative evaluations of it. Then, we show two directions of work making use of our insights to perform efficient Transfer Reinforcement Learning. The first of these directions uses Cross-Task Co-Learning and Plan Conditioned Behavioral Cloning to share skill representations produced by a Large Language Model, and it able to learn many tasks from a single demonstration each in a simulated environment. The second of these directions uses Two-Phase KL Penalization to enforce a (potentially off-policy) trust region. These advances in Transfer RL may enable robots to be used in a wider range of applications, and may also inform applying Transfer RL outside of robotics.
■590 ▼aSchool code: 0208.
■650 4▼aComputer science
■650 4▼aRobotics
■650 4▼aInformation technology
■653 ▼aImitation learning
■653 ▼aLarge Language Models
■653 ▼aMachine learning
■653 ▼aReinforcement learning
■653 ▼aTransfer learning
■690 ▼a0984
■690 ▼a0771
■690 ▼a0489
■690 ▼a0800
■71020▼aUniversity of Southern California▼bComputer Science.
■7730 ▼tDissertations Abstracts International▼g85-12B.
■790 ▼a0208
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17162492▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
