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 2019-03-06 10:00
Effects of Batch-normalization on Fisher Information Matrix of ResNet
Yasutaka Furusho, Kazushi Ikeda (NAIST) IBISML2018-110
Abstract (in Japanese) (See Japanese page) 
(in English) ResNet have intensively been studied and many techniques have been used for better performance.
Batch-normalization (BN) is one of these techniques, which accelerates convergence of training ResNet.
To clear its acceleration mechanism, we analyzed how BN affects the loss landscape, in particular, the eigenvalues of the Fisher information matrix of ResNet.
Our results show that the ResNet has eigenvalues that grow exponentially with its depth,
and that the ResNet with BN has eigenvalues that grow sub-linear with its depth.
These imply that BN allows the ResNet to use larger learning rate and hence converges faster than the vanilla ResNet.
Keyword (in Japanese) (See Japanese page) 
(in English) Neural networks / ResNet / Batch-normalization / Fisher information matrix / / / /  
Reference Info. IEICE Tech. Rep., vol. 118, no. 472, IBISML2018-110, pp. 39-44, March 2019.
Paper # IBISML2018-110 
Date of Issue 2019-02-26 (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 IBISML2018-110

Conference Information
Committee IBISML  
Conference Date 2019-03-05 - 2019-03-06 
Place (in Japanese) (See Japanese page) 
Place (in English) RIKEN AIP 
Topics (in Japanese) (See Japanese page) 
Topics (in English) Machine learning, etc. 
Paper Information
Registration To IBISML 
Conference Code 2019-03-IBISML 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Effects of Batch-normalization on Fisher Information Matrix of ResNet 
Sub Title (in English)  
Keyword(1) Neural networks  
Keyword(2) ResNet  
Keyword(3) Batch-normalization  
Keyword(4) Fisher information matrix  
Keyword(5)  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Yasutaka Furusho  
1st Author's Affiliation Nara Institute of Science and Technology (NAIST)
2nd Author's Name Kazushi Ikeda  
2nd Author's Affiliation Nara Institute of Science and Technology (NAIST)
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 ()
Speaker
Date Time 2019-03-06 10:00:00 
Presentation Time 30 
Registration for IBISML 
Paper # IEICE-IBISML2018-110 
Volume (vol) IEICE-118 
Number (no) no.472 
Page pp.39-44 
#Pages IEICE-6 
Date of Issue IEICE-IBISML-2019-02-26 


[Return to Top Page]

[Return to IEICE Web Page]


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