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 2025-07-07 13:25
Balancing Unlabeled Image Datasets by Text-based Specification of Sensitive Attributes
Kosei Suyama, Kazuaki Nakamura (TUS) ISEC2025-19 SITE2025-9 BioX2025-33 HWS2025-8 ICSS2025-19 EMM2025-22
Abstract (in Japanese) (See Japanese page) 
(in English) Modern AI models sometimes output biased results on sensitive attributes (SA) such as race and age, which causes a social issue. A reason for the biased outputs of AI models is a class imbalance problem in their training dataset. Therefore, balancing a given dataset is important. Although the class imbalance could be implicitly contained in unlabeled datasets that have no class labels, only a few existing studies focus on the implicit class imbalance problem of unlabeled datasets. In particular, no existing studies propose a method to balance unlabeled datasets where SA can be explicitly specified. In this paper, we propose a balancing method for unlabeled datasets that can explicitly specify SA by texts. The proposed method first extracts a SA-related visual feature from each image in the target imbalanced dataset using a vision-language model called
CLIP, and utilizes the features to cluster the target dataset. Then, we train a conditional diffusion model, regarding the cluster IDs as pseudo labels, and generate the same number of images from each pseudo label to construct a balanced dataset. The results of our experiments focusing on two types of SA, “race” and “age”, demonstrated that the proposed method is capable of effectively balancing a given unlabeled face dataset.
Keyword (in Japanese) (See Japanese page) 
(in English) unlabeled datasets / implicit class imbalance / sensitive attributes / conditional diffusion model / CLIP / / /  
Reference Info. IEICE Tech. Rep., vol. 125, no. 104, EMM2025-22, pp. 20-25, July 2025.
Paper # EMM2025-22 
Date of Issue 2025-06-30 (ISEC, SITE, BioX, HWS, ICSS, EMM) 
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 ISEC2025-19 SITE2025-9 BioX2025-33 HWS2025-8 ICSS2025-19 EMM2025-22

Conference Information
Committee HWS BioX ISEC SITE ICSS EMM IPSJ-CSEC IPSJ-SPT 
Conference Date 2025-07-07 - 2025-07-09 
Place (in Japanese) (See Japanese page) 
Place (in English) Sapporo Convention Center 
Topics (in Japanese) (See Japanese page) 
Topics (in English)  
Paper Information
Registration To EMM 
Conference Code 2025-07-HWS-BioX-ISEC-SITE-ICSS-EMM-CSEC-SPT 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Balancing Unlabeled Image Datasets by Text-based Specification of Sensitive Attributes 
Sub Title (in English)  
Keyword(1) unlabeled datasets  
Keyword(2) implicit class imbalance  
Keyword(3) sensitive attributes  
Keyword(4) conditional diffusion model  
Keyword(5) CLIP  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Kosei Suyama  
1st Author's Affiliation Tokyo University of Science (TUS)
2nd Author's Name Kazuaki Nakamura  
2nd Author's Affiliation Tokyo University of Science (TUS)
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 2025-07-07 13:25:00 
Presentation Time 25 minutes 
Registration for EMM 
Paper # ISEC2025-19, SITE2025-9, BioX2025-33, HWS2025-8, ICSS2025-19, EMM2025-22 
Volume (vol) vol.125 
Number (no) no.99(ISEC), no.100(SITE), no.101(BioX), no.102(HWS), no.103(ICSS), no.104(EMM) 
Page pp.20-25 
#Pages 6 
Date of Issue 2025-06-30 (ISEC, SITE, BioX, HWS, ICSS, EMM) 

[Return to Top Page]

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