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 2024-05-16 15:10
A Proposal for Generating Incorrect Pairs for the CLIP Learning based on Image Clustering
Rina Tagami, Hiroki Kobayashi, Shuichi Akizuki, Manabu Hashimoto (Chukyo Univ.) PRMU2024-5
Abstract (in Japanese) (See Japanese page) 
(in English) We proposed a novel method for generating incorrect pairs to enhance learning performance in CLIP's contrastive learning, addressing the bias in data quantity between correct and incorrect pairs. Vanilla CLIP methods that similar images are considered incorrect pairs, adversely affecting learning. Our approach extracts features of images and texts from correct pairs using ViT and BERT, clusters them, and then swaps texts of low-similarity pairs across different clusters. Experiments on the Amazon review dataset showed a 13.0% improvement in Rank@1 score over conventional CLIP and a 6.0% improvement over randomly reducing incorrect pairs.
Keyword (in Japanese) (See Japanese page) 
(in English) Large Language Model / Fine-tuning / Contrastive Learning / CLIP / Image Retrieval / / /  
Reference Info. IEICE Tech. Rep., vol. 124, no. 23, PRMU2024-5, pp. 24-29, May 2024.
Paper # PRMU2024-5 
Date of Issue 2024-05-08 (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 PRMU2024-5

Conference Information
Committee PRMU IPSJ-CVIM  
Conference Date 2024-05-15 - 2024-05-16 
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 2024-05-PRMU-CVIM 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) A Proposal for Generating Incorrect Pairs for the CLIP Learning based on Image Clustering 
Sub Title (in English)  
Keyword(1) Large Language Model  
Keyword(2) Fine-tuning  
Keyword(3) Contrastive Learning  
Keyword(4) CLIP  
Keyword(5) Image Retrieval  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Rina Tagami  
1st Author's Affiliation Chukyo University (Chukyo Univ.)
2nd Author's Name Hiroki Kobayashi  
2nd Author's Affiliation Chukyo University (Chukyo Univ.)
3rd Author's Name Shuichi Akizuki  
3rd Author's Affiliation Chukyo University (Chukyo Univ.)
4th Author's Name Manabu Hashimoto  
4th Author's Affiliation Chukyo University (Chukyo Univ.)
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 2024-05-16 15:10:00 
Presentation Time 15 minutes 
Registration for PRMU 
Paper # PRMU2024-5 
Volume (vol) vol.124 
Number (no) no.23 
Page pp.24-29 
#Pages 6 
Date of Issue 2024-05-08 (PRMU) 

[Return to Top Page]

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