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 10:30
Wider neural networks with ReLU activation generalize better
Yasutaka Furusho, Kazushi Ikeda (NAIST) IBISML2018-111
Abstract (in Japanese) (See Japanese page) 
(in English) Model size determination is important in machine learning since a larger model leads to overfitting, that is, a small training loss and a high test loss. Surprisingly, a multi-layer perceptron (MLP) with the ReLU activation function has a smaller training loss as well as a smaller gap between the training loss and a test loss as each layer gets wide.
To elucidate the mechanism, this paper theoretically derives an upper-bound of the gap for a random network and shows a wider MLP has a smaller gap.
Our numerical experiments confirmed the validity of our analysis and the applicability to the trained model.
Keyword (in Japanese) (See Japanese page) 
(in English) Neural networks / model size / over-fitting / generalization ability / / / /  
Reference Info. IEICE Tech. Rep., vol. 118, no. 472, IBISML2018-111, pp. 45-50, March 2019.
Paper # IBISML2018-111 
Date of Issue 2019-02-26 (IBISML) 
ISSN 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 IBISML2018-111

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) Wider neural networks with ReLU activation generalize better 
Sub Title (in English)  
Keyword(1) Neural networks  
Keyword(2) model size  
Keyword(3) over-fitting  
Keyword(4) generalization ability  
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 Author-1 
Date Time 2019-03-06 10:30:00 
Presentation Time 30 minutes 
Registration for IBISML 
Paper # IBISML2018-111 
Volume (vol) vol.118 
Number (no) no.472 
Page pp.45-50 
#Pages
Date of Issue 2019-02-26 (IBISML) 


[Return to Top Page]

[Return to IEICE Web Page]


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