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 2018-12-13 14:55
Fast Distributional Smoothing for CTC-VAT and its Application to Text Line Recognition
Ryohei Tanaka, Soichiro Ono, Akio Furuhata (Toshiba Digital Solutions) PRMU2018-80
Abstract (in Japanese) (See Japanese page) 
(in English) Virtual Adversarial Training (VAT), which smooths posterior distribution by minimizing distributional distance of posterior probabilities between training data and their neighborhoods, achieves success in semi-supervised learning. However, it is difficult to apply VAT to sequential label predictions such as speech recognition and text line recognition because there are so many possible label sequences that calculating posterior distributions and the distributional distance between them is costly. In this research, we propose a fast distributional smoothing method which minimizes an upper bound of the distributional distance. Furthermore, an experiment on text line recognition showed that VAT with fast distributional smoothing improved prediction accuracies by approximately 35% compared to simple self-training.
Keyword (in Japanese) (See Japanese page) 
(in English) Connectionist Temporal Classification / Virtual Adversarial Training / Semi-supervised Learning / Text Line Recognition / / / /  
Reference Info. IEICE Tech. Rep., vol. 118, no. 362, PRMU2018-80, pp. 29-34, Dec. 2018.
Paper # PRMU2018-80 
Date of Issue 2018-12-06 (PRMU) 
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 PRMU2018-80

Conference Information
Committee PRMU  
Conference Date 2018-12-13 - 2018-12-14 
Place (in Japanese) (See Japanese page) 
Place (in English)  
Topics (in Japanese) (See Japanese page) 
Topics (in English)  
Paper Information
Registration To PRMU 
Conference Code 2018-12-PRMU 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Fast Distributional Smoothing for CTC-VAT and its Application to Text Line Recognition 
Sub Title (in English)  
Keyword(1) Connectionist Temporal Classification  
Keyword(2) Virtual Adversarial Training  
Keyword(3) Semi-supervised Learning  
Keyword(4) Text Line Recognition  
Keyword(5)  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Ryohei Tanaka  
1st Author's Affiliation Toshiba Digital Solutions Corporation (Toshiba Digital Solutions)
2nd Author's Name Soichiro Ono  
2nd Author's Affiliation Toshiba Digital Solutions Corporation (Toshiba Digital Solutions)
3rd Author's Name Akio Furuhata  
3rd Author's Affiliation Toshiba Digital Solutions Corporation (Toshiba Digital Solutions)
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 ()
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 2018-12-13 14:55:00 
Presentation Time 15 minutes 
Registration for PRMU 
Paper # PRMU2018-80 
Volume (vol) vol.118 
Number (no) no.362 
Page pp.29-34 
#Pages 6 
Date of Issue 2018-12-06 (PRMU) 

[Return to Top Page]

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