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 2023-12-21 15:25
A linear time approximation of Wasserstein distance with word embedding selection
Sho Otao (Kyoto Univ.), Makoto Yamada (OIST) IBISML2023-38
Abstract (in Japanese) (See Japanese page) 
(in English) Wasserstein distance, which can be computed by solving the optimal transport problem, is a powerful method for measuring the distance between distributions. In the NLP community, it is referred as word mover's distance to measure dissimilarity between documents, treating documents as word distributions. One of the key challenges of Wasserstein distance is its computational cost since it needs cubic time. Although the Sinkhorn algorithm is a powerful tool to speed up to compute the Wasserstein distance, it still requires square time. Recently, a linear time approximation of the Wasserstein distance including the sliced Wasserstein and the tree-Wasserstein distance has been proposed. However, the linear time approximation method suffers when the dimensionality of input vectors is high. In this study, we propose a method to combine feature selection and tree approximation of Wasserstein distance to handle high-dimensional problems and compute dissimilarity between documents rapidly. More specifically, we concatenate multiple word embeddings and automatically select useful word embeddings from a concatenated embedding in a tree approximation of Wasserstein distance. To this end, we approximate Wasserstein distance for each word vector by tree approximation technique, and select the discriminative (i.e., large Wasserstein distance) word embeddings by solving an entropic regularized maximization problem. Through our synthetic experiments, we confirmed the efficacy of feature selection in our proposed method. Through our experiments on document classification, our proposed method outperformed the method that directly uses the concatenated embedding and achieved consistently high performance on all datasets.
Keyword (in Japanese) (See Japanese page) 
(in English) Optimal Transport / Group Feature Selection / Document Classification / Word Embedding / / / /  
Reference Info. IEICE Tech. Rep., vol. 123, no. 311, IBISML2023-38, pp. 50-57, Dec. 2023.
Paper # IBISML2023-38 
Date of Issue 2023-12-13 (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 IBISML2023-38

Conference Information
Committee IBISML  
Conference Date 2023-12-20 - 2023-12-21 
Place (in Japanese) (See Japanese page) 
Place (in English) National Institute of Informatics 
Topics (in Japanese) (See Japanese page) 
Topics (in English) machine learning, etc. 
Paper Information
Registration To IBISML 
Conference Code 2023-12-IBISML 
Language English (Japanese title is available) 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) A linear time approximation of Wasserstein distance with word embedding selection 
Sub Title (in English)  
Keyword(1) Optimal Transport  
Keyword(2) Group Feature Selection  
Keyword(3) Document Classification  
Keyword(4) Word Embedding  
Keyword(5)  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Sho Otao  
1st Author's Affiliation Kyoto University (Kyoto Univ.)
2nd Author's Name Makoto Yamada  
2nd Author's Affiliation Okinawa Institute of Science and Technology (OIST)
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 ()
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 2023-12-21 15:25:00 
Presentation Time 25 minutes 
Registration for IBISML 
Paper # IBISML2023-38 
Volume (vol) vol.123 
Number (no) no.311 
Page pp.50-57 
#Pages 8 
Date of Issue 2023-12-13 (IBISML) 

[Return to Top Page]

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