Historical Temporal Difference Learning: Some Initial Results

Hangzhou, Zhejiang, China June 20-June 24

DOI Bookmark:

http://doi.ieeecomputersociety.org/10.1109/IMSCCS.2006.231

2006 First International Multi-Sympos ...

	This Article
	PURCHASE ARTICLE: $19 PDF HTML IEEE Xplore Subscribers
	Share
	Email this Article to a friend
	Bibliographic References
	ASCII Text BibTex RefWorks Procite/RefMan/Endnote
	Add to:
	Digg Furl Spurl Blink Simpy Google Del.icio.us Y!MyWeb
	Search
	Similar Articles Articles by Hengshuai Yao Articles by Diao Dongcui Articles by Zengqi Sun

	ASCII Text	x
	Hengshuai Yao, Diao Dongcui, Zengqi Sun, "Historical Temporal Difference Learning: Some Initial Results," Computer and Computational Sciences, International Multi-Symposiums on, vol. 2, pp. 678-685, 2006 First International Multi-Symposiums on Computer and Computational Sciences, 2006.

	BibTex	x
	@article{ 10.1109/IMSCCS.2006.231, author = {Hengshuai Yao and Diao Dongcui and Zengqi Sun}, title = {Historical Temporal Difference Learning: Some Initial Results}, journal ={Computer and Computational Sciences, International Multi-Symposiums on}, volume = {2}, year = {2006}, isbn = {0-7695-2581-4}, pages = {678-685}, doi = {http://doi.ieeecomputersociety.org/10.1109/IMSCCS.2006.231}, publisher = {IEEE Computer Society}, address = {Los Alamitos, CA, USA}, }

	RefWorks Procite/RefMan/Endnote	x
	TY - CONF JO - Computer and Computational Sciences, International Multi-Symposiums on TI - Historical Temporal Difference Learning: Some Initial Results SN - 0-7695-2581-4 SP678 EP685 A1 - Hengshuai Yao, A1 - Diao Dongcui, A1 - Zengqi Sun, PY - 2006 KW - Multi-step Prediction KW - Reinforcement Learning KW - Temporal Difference Learning VL - 2 JA - Computer and Computational Sciences, International Multi-Symposiums on ER -

Hengshuai Yao, Tsinghua University

Diao Dongcui, Shandong University of Technology

Zengqi Sun, Tsinghua University

In this paper, we develop a multi-step prediction algorithm that is guaranteed to converge when using general function approximation. Besides, the new algorithm should satisfy the following requirements: First, it does not have to be faster than TD(0) in the look-up table representation; however, the new algorithm should be faster than residual gradient method. Second, the new algorithm should learn optimally.

Index Terms:

Multi-step Prediction, Reinforcement Learning, Temporal Difference Learning

Citation:

Hengshuai Yao, Diao Dongcui, Zengqi Sun, "Historical Temporal Difference Learning: Some Initial Results," imsccs, vol. 2, pp.678-685, 2006 First International Multi-Symposiums on Computer and Computational Sciences, 2006

Peer Review Notice | Give Us Feedback

Usage of this product signifies your acceptance of the Terms of Use.