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 2015-11-27 14:00
[Poster Presentation] Regularization by local distributional smoothing
Takeru Miyato, Shin-ichi Maeda, Masanori Koyama, Ken Nakae, Shin Ishii (Kyoto Univ.) IBISML2015-87
Abstract (in Japanese) (See Japanese page) 
(in English) Smoothness regularization is a popular method to decrease generalization error. We propose a novel regularization technique that encourages local distributional smoothness (LDS), a KL-distance based measure of the model's robustness against perturbation. The LDS is defined in terms of the direction to which the model distribution is most sensitive. Our technique resembles the adversarial training (Goodfellow et al., 2015), but distinguishes itself in that it determines the adversarial direction from the model distribution alone, and does not use the label information of data. The technique is hence applicable to semi-supervised training. When we applied our technique to the classification problem on permutation invariant MNIST, it eclipsed all the models that are not dependent on generative models or pre-training.
Keyword (in Japanese) (See Japanese page) 
(in English) Regularization / Supervised Learning / emi-supervised Learning / Adversarial training / / / /  
Reference Info. IEICE Tech. Rep., vol. 115, no. 323, IBISML2015-87, pp. 257-264, Nov. 2015.
Paper # IBISML2015-87 
Date of Issue 2015-11-19 (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 IBISML2015-87

Conference Information
Committee IBISML  
Conference Date 2015-11-25 - 2015-11-27 
Place (in Japanese) (See Japanese page) 
Place (in English) Epochal Tsukuba 
Topics (in Japanese) (See Japanese page) 
Topics (in English) Information-Based Induction Science Workshop (IBIS2015) 
Paper Information
Registration To IBISML 
Conference Code 2015-11-IBISML 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Regularization by local distributional smoothing 
Sub Title (in English)  
Keyword(1) Regularization  
Keyword(2) Supervised Learning  
Keyword(3) emi-supervised Learning  
Keyword(4) Adversarial training  
Keyword(5)  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Takeru Miyato  
1st Author's Affiliation Kyoto University (Kyoto Univ.)
2nd Author's Name Shin-ichi Maeda  
2nd Author's Affiliation Kyoto University (Kyoto Univ.)
3rd Author's Name Masanori Koyama  
3rd Author's Affiliation Kyoto University (Kyoto Univ.)
4th Author's Name Ken Nakae  
4th Author's Affiliation Kyoto University (Kyoto Univ.)
5th Author's Name Shin Ishii  
5th Author's Affiliation Kyoto University (Kyoto Univ.)
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 2015-11-27 14:00:00 
Presentation Time 180 minutes 
Registration for IBISML 
Paper # IBISML2015-87 
Volume (vol) vol.115 
Number (no) no.323 
Page pp.257-264 
#Pages 8 
Date of Issue 2015-11-19 (IBISML) 

[Return to Top Page]

[Return to IEICE Web Page]
The Institute of Electronics, Information and Communication Engineers (IEICE), Japan