Information: Join today and make your research activities more affordable! Technical workshop participation fees and annual registration fees are available at member rates.
Notice: [Important] Announcement of Changes to Registration Fee Payment and Manuscript Upload Procedures for IEICE Technical Meetings
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 2011-11-09 15:45
Modified Newton Approach to Policy Search
Hirotaka Hachiya (Tokyo Inst. of Tech.), Tetsuro Morimura (IBM Japan), Takaki Makino (Univ. of Tokyo), Masashi Sugiyama (Tokyo Inst. of Tech.) IBISML2011-54
Abstract (in Japanese) (See Japanese page) 
(in English) The natural policy gradient method was shown to be a useful approach to policy search in reinforcement learning. However, its potential weakness is that information on returns is not reflected in the metric of natural gradients, implying that it is not adaptive to data and thus less flexible. To overcome this, we propose to use Newton’s method which uses the Hessian of the expected return as a metric. However, the naive implementation of Newton’s method does not guarantee the Hessian to be negative definite, which causes instability on policy updates. To cope with this problem, we propose an adaptive scheme to keep the Hessian nonnegative. We demonstrate the effectiveness of our proposed method in standard reinforcement learning tasks.
Keyword (in Japanese) (See Japanese page) 
(in English) reinforcement learning / policy gradient / natural policy gradient / Newton's method / / / /  
Reference Info. IEICE Tech. Rep., vol. 111, no. 275, IBISML2011-54, pp. 79-85, Nov. 2011.
Paper # IBISML2011-54 
Date of Issue 2011-11-02 (IBISML) 
ISSN Print edition: ISSN 0913-5685    Online edition: ISSN 2432-6380
Copyright
and
reproduction		 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 IBISML2011-54

Conference Information
Committee IBISML  
Conference Date 2011-11-09 - 2011-11-11 
Place (in Japanese) (See Japanese page) 
Place (in English) Nara Womens Univ. 
Topics (in Japanese) (See Japanese page) 
Topics (in English) The 14th IBIS workshop 
Paper Information
Registration To IBISML 
Conference Code 2011-11-IBISML 
Language English 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Modified Newton Approach to Policy Search 
Sub Title (in English)  
Keyword(1) reinforcement learning  
Keyword(2) policy gradient  
Keyword(3) natural policy gradient  
Keyword(4) Newton's method  
Keyword(5)  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Hirotaka Hachiya  
1st Author's Affiliation Tokyo Institute of Technology (Tokyo Inst. of Tech.)
2nd Author's Name Tetsuro Morimura  
2nd Author's Affiliation IBM Japan (IBM Japan)
3rd Author's Name Takaki Makino  
3rd Author's Affiliation University of Tokyo (Univ. of Tokyo)
4th Author's Name Masashi Sugiyama  
4th Author's Affiliation Tokyo Institute of Technology (Tokyo Inst. of Tech.)
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 ()
21st Author's Name  
21st Author's Affiliation ()
22nd Author's Name  
22nd Author's Affiliation ()
23rd Author's Name  
23rd Author's Affiliation ()
24th Author's Name  
24th Author's Affiliation ()
25th Author's Name  
25th Author's Affiliation ()
26th Author's Name / /
26th Author's Affiliation ()
()
27th Author's Name / /
27th Author's Affiliation ()
()
28th Author's Name / /
28th Author's Affiliation ()
()
29th Author's Name / /
29th Author's Affiliation ()
()
30th Author's Name / /
30th Author's Affiliation ()
()
31st Author's Name / /
31st Author's Affiliation ()
()
32nd Author's Name / /
32nd Author's Affiliation ()
()
33rd Author's Name / /
33rd Author's Affiliation ()
()
34th Author's Name / /
34th Author's Affiliation ()
()
35th Author's Name / /
35th Author's Affiliation ()
()
36th Author's Name / /
36th Author's Affiliation ()
()
Speaker Author-1 
Date Time 2011-11-09 15:45:00 
Presentation Time 180 minutes 
Registration for IBISML 
Paper # IBISML2011-54 
Volume (vol) vol.111 
Number (no) no.275 
Page pp.79-85 
#Pages 7 
Date of Issue 2011-11-02 (IBISML) 

[Return to Top Page]

[Return to IEICE Web Page]
The Institute of Electronics, Information and Communication Engineers (IEICE), Japan