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 2026-03-04 13:25
Security Evaluation of Adversarial Inputs for Vision-Language Models on Edge Devices
Ryo Kumagai, Shu Takemoto, Yusuke Nozaki, Masaya Yoshikawa (Meijo Univ.) VLD2025-77 HWS2025-73 ICD2025-88
Abstract (in Japanese) (See Japanese page) 
(in English) Among artificial intelligence technologies that utilize deep learning, visual language models (VLMs) are multimodal generative models that integrate and process visual and linguistic information. Prompt injection is an attack that causes an LLM or VLM to produce incorrect output by issuing malicious instructions, causing the model to ignore safety mechanisms and instructions set by the developer. Visual prompt injection (VPI) is a particular attack targeting the visual input of a VLM, and previous research has shown that many models are vulnerable to this. However, some VPI methods exploit the high performance of large-scale models to carry out attacks, and evaluation of lightweight, small-scale models that can run on edge devices has been insufficient. In this study, we evaluate this threat by physically performing VPI on a VLM implemented on an edge device. Experimental results demonstrated that small-scale models on edge devices behave differently in response to VPI than large-scale models.
Keyword (in Japanese) (See Japanese page) 
(in English) deep learning / edge AI / vision-language models / prompt injection / M5Stack / / /  
Reference Info. IEICE Tech. Rep., vol. 125, no. 383, HWS2025-73, pp. 7-12, March 2026.
Paper # HWS2025-73 
Date of Issue 2026-02-25 (VLD, HWS, ICD) 
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 VLD2025-77 HWS2025-73 ICD2025-88

Conference Information
Committee ICD HWS VLD  
Conference Date 2026-03-04 - 2026-03-07 
Place (in Japanese) (See Japanese page) 
Place (in English)  
Topics (in Japanese) (See Japanese page) 
Topics (in English)  
Paper Information
Registration To HWS 
Conference Code 2026-03-ICD-HWS-VLD 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Security Evaluation of Adversarial Inputs for Vision-Language Models on Edge Devices 
Sub Title (in English)  
Keyword(1) deep learning  
Keyword(2) edge AI  
Keyword(3) vision-language models  
Keyword(4) prompt injection  
Keyword(5) M5Stack  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Ryo Kumagai  
1st Author's Affiliation Meijo University (Meijo Univ.)
2nd Author's Name Shu Takemoto  
2nd Author's Affiliation Meijo University (Meijo Univ.)
3rd Author's Name Yusuke Nozaki  
3rd Author's Affiliation Meijo University (Meijo Univ.)
4th Author's Name Masaya Yoshikawa  
4th Author's Affiliation Meijo University (Meijo 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 2026-03-04 13:25:00 
Presentation Time 25 minutes 
Registration for HWS 
Paper # VLD2025-77, HWS2025-73, ICD2025-88 
Volume (vol) vol.125 
Number (no) no.382(VLD), no.383(HWS), no.384(ICD) 
Page pp.7-12 
#Pages 6 
Date of Issue 2026-02-25 (VLD, HWS, ICD) 

[Return to Top Page]

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