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 2016-11-16 15:00
Statistical Mechanical Analysis of Fast Online Learning with Weight Normalization
Yuki Yoshida, Ryo Karakida, Masato Okada (UTokyo), Shun-ichi Amari (RIKEN) IBISML2016-60
Abstract (in Japanese) (See Japanese page) 
(in English) Weight normalization (WN), a newly developed optimization algorithm for neural networks by Salimans & Kingma(2016), factorizes the weight vector of a neural network into a radial length and a direction vector, and the factorized parameters follow their steepest gradient descent update. They showed that learning with WN yields better converging speed in several practical tasks including image recognition and reinforcement learning than learning with the conventional steepest descent. However, it remains theoretically unclear why this method works well. In this study, we used a statistical mechanical approach to analyze on-line learning in single layer linear and nonlinear perceptrons with WN. By deriving order parameters of the dynamics of learning, we confirmed quantitatively that WN achieves fast converging speed by automatically tuning the effective learning rate, irrespective of the nonlinearity of the neural network. This fast converging is realized when the initial value of the radial length is near the global minimum; therefore, our theory suggests that it is important to choose the initial value of the radial length appropriately when using WN.
Keyword (in Japanese) (See Japanese page) 
(in English) Neural network / Weight normalization / Online learning / Statistical mechanics / / / /  
Reference Info. IEICE Tech. Rep., vol. 116, no. 300, IBISML2016-60, pp. 101-108, Nov. 2016.
Paper # IBISML2016-60 
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 IBISML2016-60

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) Statistical Mechanical Analysis of Fast Online Learning with Weight Normalization 
Sub Title (in English)  
Keyword(1) Neural network  
Keyword(2) Weight normalization  
Keyword(3) Online learning  
Keyword(4) Statistical mechanics  
Keyword(5)  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Yuki Yoshida  
1st Author's Affiliation The University of Tokyo (UTokyo)
2nd Author's Name Ryo Karakida  
2nd Author's Affiliation The University of Tokyo (UTokyo)
3rd Author's Name Masato Okada  
3rd Author's Affiliation The University of Tokyo (UTokyo)
4th Author's Name Shun-ichi Amari  
4th Author's Affiliation RIKEN (RIKEN)
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 2016-11-16 15:00:00 
Presentation Time 180 minutes 
Registration for IBISML 
Paper # IBISML2016-60 
Volume (vol) vol.116 
Number (no) no.300 
Page pp.101-108 
#Pages 8 
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