語系:
繁體中文
English
說明(常見問題)
回圖書館首頁
手機版館藏查詢
登入
回首頁
切換:
標籤
|
MARC模式
|
ISBD
Primal-Dual Method for Reinforcement...
~
Gong, Hao.
FindBook
Google Book
Amazon
博客來
Primal-Dual Method for Reinforcement Learning and Markov Decision Processes.
紀錄類型:
書目-電子資源 : Monograph/item
正題名/作者:
Primal-Dual Method for Reinforcement Learning and Markov Decision Processes./
作者:
Gong, Hao.
出版者:
Ann Arbor : ProQuest Dissertations & Theses, : 2021,
面頁冊數:
135 p.
附註:
Source: Dissertations Abstracts International, Volume: 82-12, Section: B.
Contained By:
Dissertations Abstracts International82-12B.
標題:
Operations research. -
電子資源:
https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28494682
ISBN:
9798515256005
Primal-Dual Method for Reinforcement Learning and Markov Decision Processes.
Gong, Hao.
Primal-Dual Method for Reinforcement Learning and Markov Decision Processes.
- Ann Arbor : ProQuest Dissertations & Theses, 2021 - 135 p.
Source: Dissertations Abstracts International, Volume: 82-12, Section: B.
Thesis (Ph.D.)--Princeton University, 2021.
This item must not be sold to any third party vendors.
In this thesis, we study the application of a class of stochastic primal-dual algorithms to reinforcement learning. The method takes advantage of the min-max formulation of the Bellman equation and jointly learns the optimal value function and the optimal policy through a series of primal-dual updates. We study both the online reinforcement learning problem and the simulation problem, and present several variants of the method and their theoretical guarantees.In Chapter 2, we investigate the Infinite-Horizon Average-Reward Markov Decision Process (AMDP) and propose an online primal-dual algorithm to learn the optimal policy by actively making decisions along a single state trajectory. Our results appear to be the first duality-based value-policy gradient method and regret analysis for infinite-horizon RL. We prove an \uD835\uDCDE(√T) regret upper bound that also depends on an ergodicity parameter of the AMDP, which can be significantly smaller than existing diameter-dependent bounds in some cases.In Chapter 3, we propose a class of utility maximization problems as a generalization of AMDPs that incorporates nonlinearity into the objective. We develop a primal-dual algorithm to solve the proposed problems and show that it achieves superior sample complexity compared to previous duality-based algorithms when the problem reduces to an AMDP.Chapter 4 presents a general primal-dual framework for solving the utility maximization problem with function approximation. We also study a specific class of linear/bilinear models and prove that the adjusted algorithm achieves a comparable convergence rate.
ISBN: 9798515256005Subjects--Topical Terms:
547123
Operations research.
Subjects--Index Terms:
Machine Learning
Primal-Dual Method for Reinforcement Learning and Markov Decision Processes.
LDR
:02969nmm a2200421 4500
001
2282952
005
20211022115807.5
008
220723s2021 ||||||||||||||||| ||eng d
020
$a
9798515256005
035
$a
(MiAaPQ)AAI28494682
035
$a
AAI28494682
040
$a
MiAaPQ
$c
MiAaPQ
100
1
$a
Gong, Hao.
$3
3561838
245
1 0
$a
Primal-Dual Method for Reinforcement Learning and Markov Decision Processes.
260
1
$a
Ann Arbor :
$b
ProQuest Dissertations & Theses,
$c
2021
300
$a
135 p.
500
$a
Source: Dissertations Abstracts International, Volume: 82-12, Section: B.
500
$a
Advisor: Wang, Mengdi.
502
$a
Thesis (Ph.D.)--Princeton University, 2021.
506
$a
This item must not be sold to any third party vendors.
520
$a
In this thesis, we study the application of a class of stochastic primal-dual algorithms to reinforcement learning. The method takes advantage of the min-max formulation of the Bellman equation and jointly learns the optimal value function and the optimal policy through a series of primal-dual updates. We study both the online reinforcement learning problem and the simulation problem, and present several variants of the method and their theoretical guarantees.In Chapter 2, we investigate the Infinite-Horizon Average-Reward Markov Decision Process (AMDP) and propose an online primal-dual algorithm to learn the optimal policy by actively making decisions along a single state trajectory. Our results appear to be the first duality-based value-policy gradient method and regret analysis for infinite-horizon RL. We prove an \uD835\uDCDE(√T) regret upper bound that also depends on an ergodicity parameter of the AMDP, which can be significantly smaller than existing diameter-dependent bounds in some cases.In Chapter 3, we propose a class of utility maximization problems as a generalization of AMDPs that incorporates nonlinearity into the objective. We develop a primal-dual algorithm to solve the proposed problems and show that it achieves superior sample complexity compared to previous duality-based algorithms when the problem reduces to an AMDP.Chapter 4 presents a general primal-dual framework for solving the utility maximization problem with function approximation. We also study a specific class of linear/bilinear models and prove that the adjusted algorithm achieves a comparable convergence rate.
590
$a
School code: 0181.
650
4
$a
Operations research.
$3
547123
650
4
$a
Computer science.
$3
523869
650
4
$a
Applied mathematics.
$3
2122814
650
4
$a
Artificial intelligence.
$3
516317
653
$a
Machine Learning
653
$a
Stochastic primal-dual algorithms
653
$a
Bellman equation
653
$a
Optimal value function
653
$a
Optimal policy
653
$a
Theoretical guarantees
653
$a
Utility maximization
653
$a
Infinite-Horizon Average-Reward Markov Decision Process
690
$a
0796
690
$a
0984
690
$a
0800
690
$a
0364
710
2
$a
Princeton University.
$b
Operations Research and Financial Engineering.
$3
2096743
773
0
$t
Dissertations Abstracts International
$g
82-12B.
790
$a
0181
791
$a
Ph.D.
792
$a
2021
793
$a
English
856
4 0
$u
https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28494682
筆 0 讀者評論
館藏地:
全部
電子資源
出版年:
卷號:
館藏
1 筆 • 頁數 1 •
1
條碼號
典藏地名稱
館藏流通類別
資料類型
索書號
使用類型
借閱狀態
預約狀態
備註欄
附件
W9434685
電子資源
11.線上閱覽_V
電子書
EB
一般使用(Normal)
在架
0
1 筆 • 頁數 1 •
1
多媒體
評論
新增評論
分享你的心得
Export
取書館
處理中
...
變更密碼
登入