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 2016-01-29 16:15
Proposal of novel dropout method and its analysis of dynamic property
Daisuke Saitoh, Tasuku Kondo, Kazuyuki Hara (Nihon Univ.) NC2015-67
Abstract (in Japanese) (See Japanese page) 
(in English) Deep learning that use a large network and includes many units tends to occur the overfitting. Therefore, to avoid the overfitting, several regularization methods have been proposed, and one of them is Dropout. Dropout selects some units of the network at random and it drops them in the learning process. Then the network size becomes smaller, and the overfitting can be avoided. When calculating the output, we sum up the dropped units and that of not dropped units. This seems like the ensemble learning. On the other hand, there is a phenomenon called the symmetry breaking that achieves very small residual error. Then, we treated Dropout as the ensemble learning, and explored the Dropout that achieves the symmetry breaking.
Keyword (in Japanese) (See Japanese page) 
(in English) Dropout / regularization / online learning / neural networks / / / /  
Reference Info. IEICE Tech. Rep., vol. 115, no. 426, NC2015-67, pp. 55-60, Jan. 2016.
Paper # NC2015-67 
Date of Issue 2016-01-21 (NC) 
ISSN Print edition: ISSN 0913-5685    Online edition: ISSN 2432-6380
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 NC2015-67

Conference Information
Committee NC NLP  
Conference Date 2016-01-28 - 2016-01-29 
Place (in Japanese) (See Japanese page) 
Place (in English) Kyushu Institute of Technology 
Topics (in Japanese) (See Japanese page) 
Topics (in English) Implementation of Neuro Computing,Analysis and Modeling of Human Science, etc 
Paper Information
Registration To NC 
Conference Code 2016-01-NC-NLP 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Proposal of novel dropout method and its analysis of dynamic property 
Sub Title (in English)  
Keyword(1) Dropout  
Keyword(2) regularization  
Keyword(3) online learning  
Keyword(4) neural networks  
1st Author's Name Daisuke Saitoh  
1st Author's Affiliation Nihon University (Nihon Univ.)
2nd Author's Name Tasuku Kondo  
2nd Author's Affiliation Nihon University (Nihon Univ.)
3rd Author's Name Kazuyuki Hara  
3rd Author's Affiliation Nihon University (Nihon 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 2016-01-29 16:15:00 
Presentation Time 25 minutes 
Registration for NC 
Paper # NC2015-67 
Volume (vol) vol.115 
Number (no) no.426 
Page pp.55-60 
Date of Issue 2016-01-21 (NC) 

[Return to Top Page]

[Return to IEICE Web Page]

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