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 2016-11-17 14:00
Incremental Natural Actor Critic with Importance Weight Aware Update
Ryo Iwaki (Osaka Univ.), Hiroki Yokoyama (Tamagawa Univ.), Minoru Asada (Osaka Univ.) IBISML2016-81
Abstract (in Japanese) (See Japanese page) 
(in English) Appropriate tuning of step-size parameter is crucial for reinforcement learning, as well as other machine learning techniques. For linear regression cases, importance weight aware update (IWAU) guarantees a safety; the updated parameter will not overshoot the stationary point of loss function. If a squared loss is utilized, IWAU can be derived as a simple closed form. In this paper, We focus on the natural policy gradient (NPG) methods, which can be derived as linear function approximation and squared loss, and derive new adaptive step-size for NPG estimation method, which is safe and free from step-size meta-parameter tuning. We confirm the usefulness of the proposed step-size in classical benchmark for natural policy gradient method.
Keyword (in Japanese) (See Japanese page) 
(in English) reinforcement learning / natural policy gradient method / adaptive step-size / incremental learning / importance weight aware update / / /  
Reference Info. IEICE Tech. Rep., vol. 116, no. 300, IBISML2016-81, pp. 251-257, Nov. 2016.
Paper # IBISML2016-81 
Date of Issue 2016-11-09 (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 IBISML2016-81

Conference Information
Committee IBISML  
Conference Date 2016-11-16 - 2016-11-18 
Place (in Japanese) (See Japanese page) 
Place (in English) Kyoto Univ. 
Topics (in Japanese) (See Japanese page) 
Topics (in English) Information-Based Induction Science Workshop (IBIS2016) 
Paper Information
Registration To IBISML 
Conference Code 2016-11-IBISML 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Incremental Natural Actor Critic with Importance Weight Aware Update 
Sub Title (in English)  
Keyword(1) reinforcement learning  
Keyword(2) natural policy gradient method  
Keyword(3) adaptive step-size  
Keyword(4) incremental learning  
Keyword(5) importance weight aware update  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Ryo Iwaki  
1st Author's Affiliation Osaka University (Osaka Univ.)
2nd Author's Name Hiroki Yokoyama  
2nd Author's Affiliation Tamagawa University (Tamagawa Univ.)
3rd Author's Name Minoru Asada  
3rd Author's Affiliation Osaka University (Osaka Univ.)
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 ()
Speaker Author-1 
Date Time 2016-11-17 14:00:00 
Presentation Time 180 minutes 
Registration for IBISML 
Paper # IBISML2016-81 
Volume (vol) vol.116 
Number (no) no.300 
Page pp.251-257 
#Pages
Date of Issue 2016-11-09 (IBISML) 


[Return to Top Page]

[Return to IEICE Web Page]


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