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 2026-03-05 15:00
[Poster Presentation] Defense against Model Extraction Attacks on Diffusion Models by Embedding Inheritable Watermarks into Generated Images
Hayato Shoji, Kazuaki Nakamura (Tokyo Univ. of Science) EMM2025-125
Abstract (in Japanese) (See Japanese page) 
(in English) In recent years, image generation models have become widely adopted and have achieved remarkable success in various fields, including content creation. However, the threat of Model Extraction Attack (MEA) has been increasing, in which an adversary issues a large number of queries to a target model and uses the resulting input–output pairs as training data to illicitly construct a clone model with functionality comparable to that of the original model. Since attackers can potentially obtain substantial profits by creating and releasing such clone models, the threat of MEA targeting image generation models is rapidly increasing. To address this threat, we propose a defense method by detecting such illegally constructed clone models after their creation. Specifically, we embed a watermark into images generated by the original model in a manner that is inherited by images generated by the clone model. In detection, we determine whether a target model is a clone by comparing a set of images generated by the inspected model with a set of images generated by a shadow model trained using the watermarked images. Experimental results demonstrate that the proposed method is effective for detecting clone models and exhibits high robustness against attacks involving disruptive operations and data augmentations, such as gaussian noise, JPEG compression, horizontal flipping, and image resizing.
Keyword (in Japanese) (See Japanese page) 
(in English) Model Extraction Attack / image generation model / security / / / / /  
Reference Info. IEICE Tech. Rep., vol. 125, no. 396, EMM2025-125, pp. 47-52, March 2026.
Paper # EMM2025-125 
Date of Issue 2026-02-26 (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 EMM2025-125

Conference Information
Committee EMM  
Conference Date 2026-03-05 - 2026-03-06 
Place (in Japanese) (See Japanese page) 
Place (in English) Miyako Island Hirara Port Terminal Bldg. 
Topics (in Japanese) (See Japanese page) 
Topics (in English)  
Paper Information
Registration To EMM 
Conference Code 2026-03-EMM 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Defense against Model Extraction Attacks on Diffusion Models by Embedding Inheritable Watermarks into Generated Images 
Sub Title (in English)  
Keyword(1) Model Extraction Attack  
Keyword(2) image generation model  
Keyword(3) security  
Keyword(4)  
Keyword(5)  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Hayato Shoji  
1st Author's Affiliation Tokyo University of Science (Tokyo Univ. of Science)
2nd Author's Name Kazuaki Nakamura  
2nd Author's Affiliation Tokyo University of Science (Tokyo Univ. of Science)
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 2026-03-05 15:00:00 
Presentation Time 60 minutes 
Registration for EMM 
Paper # EMM2025-125 
Volume (vol) vol.125 
Number (no) no.396 
Page pp.47-52 
#Pages 6 
Date of Issue 2026-02-26 (EMM) 

[Return to Top Page]

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