RUcore: Rutgers University Community Repository
A unifying framework for computational reinforcement learning theory
Descriptive
TypeOfResource
Text
TitleInfo
(ID = T-1)
Title
A unifying framework for computational reinforcement learning theory
SubTitle
PartName
PartNumber
NonSort
Identifier
(displayLabel = );
(invalid = )
ETD_1766
Identifier
(type = hdl)
http://hdl.rutgers.edu/1782.2/rucore10001600001.ETD.000051858
Language
(objectPart = )
LanguageTerm
(authority = ISO639-2);
(type = code)
eng
Genre
(authority = marcgt)
theses
Subject
(ID = SBJ-1);
(authority = RUETD)
Topic
Computer Science
Subject
(ID = SBJ-2);
(authority = ETD-LCSH)
Topic
Computational learning theory
Subject
(ID = SBJ-3);
(authority = ETD-LCSH)
Topic
Reinforcement learning
Subject
(ID = SBJ-4);
(authority = ETD-LCSH)
Topic
Machine learning
Abstract
Computational learning theory studies mathematical models that allow one to formally analyze and compare the performance of supervised-learning algorithms such as their sample complexity. While existing models such as PAC (Probably Approximately Correct) have played an influential role in understanding the nature of supervised learning, they have not been as successful in reinforcement learning (RL). Here, the fundamental barrier is the need for active exploration in sequential decision problems.
An RL agent tries to maximize long-term utility by exploiting its knowledge about the problem, but this knowledge has to be acquired by the agent itself through exploring the problem that may reduce short-term utility. The need for active exploration is common in many problems in daily life, engineering, and sciences. For example, a Backgammon program strives to take good moves to maximize the probability of winning a game, but sometimes it may try novel and possibly harmful moves to discover how the opponent reacts in the hope of discovering a better game-playing strategy. It has been known since the early days of RL that a good tradeoff between exploration and exploitation is critical for the agent to learn fast (i.e., to reach near-optimal strategies with a small sample complexity), but a general theoretical analysis of this tradeoff remained open until recently.
In this dissertation, we introduce a novel computational learning model called KWIK (Knows What It Knows) that is designed particularly for its utility in analyzing learning problems like RL where active exploration can impact the training data the learner is exposed to. My thesis is that the KWIK learning model provides a flexible, modularized, and unifying way for creating and analyzing reinforcement-learning algorithms with provably efficient exploration. In particular, we show how the KWIK perspective can be used to unify the analysis of existing RL algorithms with polynomial sample complexity. It also facilitates the development of new algorithms with smaller sample complexity, which have demonstrated empirically faster learning speed in real-world problems. Furthermore, we provide an improved, matching sample complexity lower bound, which suggests the optimality (in a sense) of one of the KWIK-based algorithms known as delayed Q-learning.
An RL agent tries to maximize long-term utility by exploiting its knowledge about the problem, but this knowledge has to be acquired by the agent itself through exploring the problem that may reduce short-term utility. The need for active exploration is common in many problems in daily life, engineering, and sciences. For example, a Backgammon program strives to take good moves to maximize the probability of winning a game, but sometimes it may try novel and possibly harmful moves to discover how the opponent reacts in the hope of discovering a better game-playing strategy. It has been known since the early days of RL that a good tradeoff between exploration and exploitation is critical for the agent to learn fast (i.e., to reach near-optimal strategies with a small sample complexity), but a general theoretical analysis of this tradeoff remained open until recently.
In this dissertation, we introduce a novel computational learning model called KWIK (Knows What It Knows) that is designed particularly for its utility in analyzing learning problems like RL where active exploration can impact the training data the learner is exposed to. My thesis is that the KWIK learning model provides a flexible, modularized, and unifying way for creating and analyzing reinforcement-learning algorithms with provably efficient exploration. In particular, we show how the KWIK perspective can be used to unify the analysis of existing RL algorithms with polynomial sample complexity. It also facilitates the development of new algorithms with smaller sample complexity, which have demonstrated empirically faster learning speed in real-world problems. Furthermore, we provide an improved, matching sample complexity lower bound, which suggests the optimality (in a sense) of one of the KWIK-based algorithms known as delayed Q-learning.
PhysicalDescription
Form
(authority = gmd)
electronic resource
Extent
xvii, 264 p. : ill.
InternetMediaType
application/pdf
InternetMediaType
text/xml
Note
(type = degree)
Ph.D.
Note
(type = bibliography)
Includes bibliographical references (p. 238-261)
Note
(type = statement of responsibility)
by Lihong Li
Name
(ID = NAME-1);
(type = personal)
NamePart
(type = family)
Li
NamePart
(type = given)
Lihong
NamePart
(type = date)
1979-
Role
RoleTerm
(authority = RULIB);
(type = )
author
DisplayForm
Lihong Li
Name
(ID = NAME-2);
(type = personal)
NamePart
(type = family)
Littman
NamePart
(type = given)
Michael
Role
RoleTerm
(authority = RULIB);
(type = )
chair
Affiliation
Advisory Committee
DisplayForm
Michael L. Littman
Name
(ID = NAME-3);
(type = personal)
NamePart
(type = family)
Pazzani
NamePart
(type = given)
Michael
Role
RoleTerm
(authority = RULIB);
(type = )
internal member
Affiliation
Advisory Committee
DisplayForm
Michael J. Pazzani
Name
(ID = NAME-4);
(type = personal)
NamePart
(type = family)
Szegedy
NamePart
(type = given)
Mario
Role
RoleTerm
(authority = RULIB);
(type = )
internal member
Affiliation
Advisory Committee
DisplayForm
Mario Szegedy
Name
(ID = NAME-5);
(type = personal)
NamePart
(type = family)
Schapire
NamePart
(type = given)
Robert
Role
RoleTerm
(authority = RULIB);
(type = )
outside member
Affiliation
Advisory Committee
DisplayForm
Robert E. Schapire
Name
(ID = NAME-1);
(type = corporate)
NamePart
Rutgers University
Role
RoleTerm
(authority = RULIB);
(type = )
degree grantor
Name
(ID = NAME-2);
(type = corporate)
NamePart
Graduate School - New Brunswick
Role
RoleTerm
(authority = RULIB);
(type = )
school
OriginInfo
DateCreated
(point = );
(qualifier = exact)
2009
DateOther
(qualifier = exact);
(type = degree)
2009-10
Place
PlaceTerm
(type = code)
xx
RelatedItem
(type = host)
TitleInfo
Title
Rutgers University Electronic Theses and Dissertations
Identifier
(type = RULIB)
ETD
RelatedItem
(type = host)
TitleInfo
Title
Graduate School - New Brunswick Electronic Theses and Dissertations
Identifier
(type = local)
rucore19991600001
Location
PhysicalLocation
(authority = marcorg);
(displayLabel = Rutgers, The State University of New Jersey)
NjNbRU
Identifier
(type = doi)
doi:10.7282/T3S46S46
Genre
(authority = ExL-Esploro)
ETD doctoral
Back to the top
Rights
RightsDeclaration
(AUTHORITY = GS);
(ID = rulibRdec0006)
The author owns the copyright to this work
Copyright
Status
Copyright protected
Notice
Note
Availability
Status
Open
Reason
Permission or license
Note
RightsHolder
(ID = PRH-1);
(type = personal)
Name
FamilyName
Li
GivenName
Lihong
Role
Copyright holder
RightsEvent
(ID = RE-1);
(AUTHORITY = rulib)
Type
Permission or license
Label
Place
DateTime
Detail
AssociatedEntity
(ID = AE-1);
(AUTHORITY = rulib)
Role
Copyright holder
Name
Lihong Li
Affiliation
Rutgers University. Graduate School - New Brunswick
AssociatedObject
(ID = AO-1);
(AUTHORITY = rulib)
Type
License
Name
Author Agreement License
Detail
I hereby grant to the Rutgers University Libraries and to my school the non-exclusive right to archive, reproduce and distribute my thesis or dissertation, in whole or in part, and/or my abstract, in whole or in part, in and from an electronic format, subject to the release date subsequently stipulated in this submittal form and approved by my school. I represent and stipulate that the thesis or dissertation and its abstract are my original work, that they do not infringe or violate any rights of others, and that I make these grants as the sole owner of the rights to my thesis or dissertation and its abstract. I represent that I have obtained written permissions, when necessary, from the owner(s) of each third party copyrighted matter to be included in my thesis or dissertation and will supply copies of such upon request by my school. I acknowledge that RU ETD and my school will not distribute my thesis or dissertation or its abstract if, in their reasonable judgment, they believe all such rights have not been secured. I acknowledge that I retain ownership rights to the copyright of my work. I also retain the right to use all or part of this thesis or dissertation in future works, such as articles or books.
Back to the top
Technical
ContentModel
ETD
MimeType
(TYPE = file)
application/pdf
MimeType
(TYPE = container)
application/x-tar
FileSize
(UNIT = bytes)
2826240
Checksum
(METHOD = SHA1)
ec6186a968f05092840fa87acce903aaef59ecd6
Back to the top
Statistical Profile