IEICE Technical Committee Submission System
Conference Paper's Information
Online Proceedings
[Sign in]
Tech. Rep. Archives
 Go Top Page Go Previous   [Japanese] / [English] 

Paper Abstract and Keywords
Presentation 2007-12-22 16:10
The Policy Gradient On The Average Reward Manifold
Takamitsu Matsubara (NAIST/ATR), Jun Morimoto (JST/ATR) NC2007-85
Abstract (in Japanese) (See Japanese page) 
(in English) In this paper we propose a novel policy gradient type reinforcement learning method on the average rewardmanifold, in which a metric to measure the effect of change in policy parameters on the average reward is introduced. In our method, the derivative of the average reward with respect to the policy improvement can be fixed as a constant. Moreover,
around a (sub-) optimal policy, the policy gradient method is equivalent to the Newton method. Simple simulation results with comparison to previously proposed natural policy gradient methods demonstrate the effectiveness of our policy gradient method.
Keyword (in Japanese) (See Japanese page) 
(in English) Reinforcement Learning / Policy Gradient / Natural Policy Gradient / / / / /  
Reference Info. IEICE Tech. Rep., vol. 107, no. 410, NC2007-85, pp. 81-86, Dec. 2007.
Paper # NC2007-85 
Date of Issue 2007-12-15 (NC) 
ISSN Print edition: ISSN 0913-5685  Online edition: ISSN 2432-6380
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034)
Download PDF NC2007-85

Conference Information
Committee MBE NC  
Conference Date 2007-12-22 - 2007-12-22 
Place (in Japanese) (See Japanese page) 
Place (in English)  
Topics (in Japanese) (See Japanese page) 
Topics (in English)  
Paper Information
Registration To NC 
Conference Code 2007-12-MBE-NC 
Language English (Japanese title is available) 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) The Policy Gradient On The Average Reward Manifold 
Sub Title (in English)  
Keyword(1) Reinforcement Learning  
Keyword(2) Policy Gradient  
Keyword(3) Natural Policy Gradient  
1st Author's Name Takamitsu Matsubara  
1st Author's Affiliation Nara Institute of Science and Technology, ATR-CNS (NAIST/ATR)
2nd Author's Name Jun Morimoto  
2nd Author's Affiliation JST-ICORP, ATR-CNS (JST/ATR)
3rd Author's Name  
3rd Author's Affiliation ()
4th Author's Name  
4th Author's Affiliation ()
5th Author's Name  
5th Author's Affiliation ()
6th Author's Name  
6th Author's Affiliation ()
7th Author's Name  
7th Author's Affiliation ()
8th Author's Name  
8th Author's Affiliation ()
9th Author's Name  
9th Author's Affiliation ()
10th Author's Name  
10th Author's Affiliation ()
11th Author's Name  
11th Author's Affiliation ()
12th Author's Name  
12th Author's Affiliation ()
13th Author's Name  
13th Author's Affiliation ()
14th Author's Name  
14th Author's Affiliation ()
15th Author's Name  
15th Author's Affiliation ()
16th Author's Name  
16th Author's Affiliation ()
17th Author's Name  
17th Author's Affiliation ()
18th Author's Name  
18th Author's Affiliation ()
19th Author's Name  
19th Author's Affiliation ()
20th Author's Name  
20th Author's Affiliation ()
Date Time 2007-12-22 16:10:00 
Presentation Time 25 
Registration for NC 
Paper # IEICE-NC2007-85 
Volume (vol) IEICE-107 
Number (no) no.410 
Page pp.81-86 
#Pages IEICE-6 
Date of Issue IEICE-NC-2007-12-15 

[Return to Top Page]

[Return to IEICE Web Page]

The Institute of Electronics, Information and Communication Engineers (IEICE), Japan