- 1、本文档共25页,可阅读全部内容。
- 2、原创力文档(book118)网站文档一经付费(服务费),不意味着购买了该文档的版权,仅供个人/单位学习、研究之用,不得用于商业用途,未经授权,严禁复制、发行、汇编、翻译或者网络传播等,侵权必究。
- 3、本站所有内容均由合作方或网友上传,本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺!文档内容仅供研究参考,付费前请自行鉴别。如您付费,意味着您自己接受本站规则且自行承担风险,本站不退款、不进行额外附加服务;查看《如何避免下载的几个坑》。如果您已付费下载过本站文档,您可以点击 这里二次下载。
- 4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等,请点击“版权申诉”(推荐),也可以打举报电话:400-050-0827(电话支持时间:9:00-18:30)。
查看更多
Journal of Machine Learning Research 5 (2003) 1-25 Submitted 4/02; Revised 9/02; Published 12/03
Learning Rates for Q-learning
Eyal Even-Dar EVEND @CS .TAU .AC .IL
Yishay Mansour MANSOUR @CS .TAU .AC .IL
School of Computer Science
Tel-Aviv University
Tel-Aviv, 69978, Israel
Editor: Peter Bartlett
Abstract
In this paper we derive convergence rates for Q-learning. We show an interesting relationship
between the convergence rate and the learning rate used in Q-learning. For a polynomial learning
rate, one which is 1tω at time t where ω 12 1, we show that the convergence rate is poly-
nomial in 11 −γ, where γ is the discount factor. In contrast we show that for a linear learning
rate, one which is 1t at time t, the convergence rate has an exponential dependence on 11 −γ.
In addition we show a simple example that proves this exponential behavior is inherent for linear
learning rates.
Keywords: Reinforcement Learning, Q-Learning, Stochastic Processes, Convergence Bounds,
Learning Rates.
1. Introduction
In Reinforcement Learning, an agent wanders in an unknown environment and tries to maximize its
long term return by performing actions and receiving rewards. The challenge is to understand how
a current action will affect future rewards. A good way to model this task is with Markov Decision
Processes (MDP), which have become the dominant approach in Reinforcement Learning (Sutton
and Barto, 1998, Bertsekas and Tsitsiklis, 1996).
An MDP includes states (which abstract the environment), actions (which are the available
actions to the agent), and for each state-action pair, a distribution of
您可能关注的文档
- 《10~22MPa双色水位计》.doc
- 《11 Jan 2016 盟會資料庫.微愽版docx》.docx
- 《11-22184哈尔滨MH-FJ-3ARS说明书封包机》.pdf
- 《11.03 m3 U1》.ppt
- 《11.20 M3 U3》.ppt
- 《11.27 M3 U5》.ppt
- 《11001000中山大学管理学院2016年MBA招生简》.doc
- 《110m3沉析槽的设计》.doc
- 《11172-3 mp3协议》.pdf
- 《114_Backup QNAP NAS data to Microsoft Azure storage》.docx
- 重庆七中2025届高考仿真卷化学试卷含解析.doc
- 福建省福州市福建师大附中2025届高考全国统考预测密卷生物试卷含解析.doc
- 山东省聊城市高唐一中2025届高三下学期联考生物试题含解析.doc
- 2025届贵州省黔东南市重点中学高考历史一模试卷含解析.doc
- 2025届广东省五校高三下学期联合考试生物试题含解析.doc
- 河南省邓州市花洲实验高级中学2025届高考化学倒计时模拟卷含解析.doc
- 2025届忻州市第一中学高考压轴卷生物试卷含解析.doc
- 安徽师范大学附属中学2025届高考历史倒计时模拟卷含解析.doc
- 2025届山东省临淄中学高三最后一卷历史试卷含解析.doc
- 湖南省长沙市宁乡一中2025届高三压轴卷历史试卷含解析.doc
文档评论(0)