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 2019-01-23 15:40
Measuring the Convolution Neural Network similarities trained with different dataset using SVCCA
Toya Teramoto, Hayaru Shouno (UEC) NC2018-40
Abstract (in Japanese) (See Japanese page) 
(in English) Deep Convolutional Neural Network (DCNN) is a successful model in the field of computer vision such like image classification and segmentation and so on.

However, the DCNN has still been regarded as block box.
And many researches are try to analyze what kind of feature expression DCNN learns in recent years.

Understanding the DCNN is important because it leads to an improvement in performance, and it is also available in products where an black-box system can not be used.

In this research, in oder to analyzing, we investigate the difference of feature representations in DCNNs which are trained with similar datasets.

We also investigate the difference from the viewpoint of the model structures.

In this study, we compare models learned with different but similar datasets and consider how much difference will occur in the network by changing the capacity of the model.

As a results, we obtain the network capacity is important to keep similarity in representations.
Keyword (in Japanese) (See Japanese page) 
(in English) Deep Learning / Neocognitron / Canonical Correlation Analysis / / / / /  
Reference Info. IEICE Tech. Rep., vol. 118, no. 414, NC2018-40, pp. 11-16, Jan. 2019.
Paper # NC2018-40 
Date of Issue 2019-01-16 (NC) 
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 NC2018-40

Conference Information
Committee NLP NC  
Conference Date 2019-01-23 - 2019-01-24 
Place (in Japanese) (See Japanese page) 
Place (in English) The Centennial Hall, Hokkaido Univ. 
Topics (in Japanese) (See Japanese page) 
Topics (in English) General Implementation of Neuro Computing, Analysis and Modeling of Human Science, etc. 
Paper Information
Registration To NC 
Conference Code 2019-01-NLP-NC 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Measuring the Convolution Neural Network similarities trained with different dataset using SVCCA 
Sub Title (in English)  
Keyword(1) Deep Learning  
Keyword(2) Neocognitron  
Keyword(3) Canonical Correlation Analysis  
Keyword(4)  
Keyword(5)  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Toya Teramoto  
1st Author's Affiliation University of Electro-Communications (UEC)
2nd Author's Name Hayaru Shouno  
2nd Author's Affiliation University of Electro-Communications (UEC)
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 2019-01-23 15:40:00 
Presentation Time 20 minutes 
Registration for NC 
Paper # NC2018-40 
Volume (vol) vol.118 
Number (no) no.414 
Page pp.11-16 
#Pages
Date of Issue 2019-01-16 (NC) 


[Return to Top Page]

[Return to IEICE Web Page]


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