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 2024-06-06 10:20
The Relationship between Power Laws in Neural Representation and Image Recognition
Riku Matsumoto, Yasuhiro Tsuno (Ritsumeikan Univ.) NLP2024-16 CCS2024-3
Abstract (in Japanese) (See Japanese page) 
(in English) Recent neuroscience research has found that when examining the dimensionality of the neural state space in the primary visual cortex of mice, principal component analysis (PCA) of population neural responses to a large number of images reveals a power-law decay in the variance spectrum.
This power-law exponent suggests the nature of information representation of images in the neural state space. Concurrently, in the field of artificial neural networks, it has been suggested that this power-law exponent of the variance spectrum may be important for robustness against adversarial attacks.
In this study, we investigated the relationship between the power-law exponent of the hidden layers of models trained with the Resnet50 architecture—specifically models robust to adversarial attacks, models robust to image corruption, and standard models—and their various image recognition performance (adversarial attacks, image corruption, and standard images).
Additionally, it was found that the power-law exponent changes when the stochasticity of neurons in probabilistic artificial neural networks, which can replicate the same power-law exponent as mice, is varied.
Therefore, we investigated the relationship between the power-law exponent and image recognition performance when the stochasticity is altered.
Keyword (in Japanese) (See Japanese page) 
(in English) Neural State Space / Power Law / Convolutional Neural Network / Adversarial Attack / Image Corruption / Stochasticity / /  
Reference Info. IEICE Tech. Rep., vol. 124, no. 63, CCS2024-3, pp. 8-13, June 2024.
Paper # CCS2024-3 
Date of Issue 2024-05-30 (NLP, CCS) 
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 NLP2024-16 CCS2024-3

Conference Information
Committee NLP CCS  
Conference Date 2024-06-06 - 2024-06-07 
Place (in Japanese) (See Japanese page) 
Place (in English) West Japan General Exhibition Center AIM 
Topics (in Japanese) (See Japanese page) 
Topics (in English) Nonlinear Problems, Complex Communication Sciences, etc. 
Paper Information
Registration To CCS 
Conference Code 2024-06-NLP-CCS 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) The Relationship between Power Laws in Neural Representation and Image Recognition 
Sub Title (in English)  
Keyword(1) Neural State Space  
Keyword(2) Power Law  
Keyword(3) Convolutional Neural Network  
Keyword(4) Adversarial Attack  
Keyword(5) Image Corruption  
Keyword(6) Stochasticity  
Keyword(7)  
Keyword(8)  
1st Author's Name Riku Matsumoto  
1st Author's Affiliation Ritsumeikan University (Ritsumeikan Univ.)
2nd Author's Name Yasuhiro Tsuno  
2nd Author's Affiliation Ritsumeikan University (Ritsumeikan Univ.)
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 2024-06-06 10:20:00 
Presentation Time 25 minutes 
Registration for CCS 
Paper # NLP2024-16, CCS2024-3 
Volume (vol) vol.124 
Number (no) no.62(NLP), no.63(CCS) 
Page pp.8-13 
#Pages 6 
Date of Issue 2024-05-30 (NLP, CCS) 

[Return to Top Page]

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