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 2024-11-12 14:05
Low-bit Quantization Methods for Neural Networks
Emi Wada, Shinji Kimuda (Waseda Univ.) VLD2024-28 ICD2024-46 DC2024-50 RECONF2024-58
Abstract (in Japanese) (See Japanese page) 
(in English) Lightweighting neural networks are critical challenges for edge applications. Among various methods, Post Training Quantization (PTQ) is a simple and effective approach for reducing the size of Deep Neural Network (DNN) models. While various PTQ techniques have been explored, logarithmic quantization methods show high performance at ultra-low precision of 4 bits or less. In this manuscipt, we propose a method to combine different logarithmic-based quantizers to weights and activations. For weight quantization, Subset Quantization (SQ) is used, which offers high flexibility in handling distribution. SQ is a quantizer that searches for optimal quantization points suitable for the data distribution from a set of logarithmic values composed of two-word addends, called the universal set. Since all possible combinations are thoroughly searched, it is possible to obtain highly accurate quantization points. For activation quantization, we apply Selective Two-word Log-scale Quantization (STLQ), which is a more computationally efficient method. By setting a two-word rate and prioritizing the logarithmic quantization of values with large quantization errors, STLQ efficiently improves accuracy. By combining different quantization techniques for weights and activations, we achieve enhanced precision. Evaluation on large-scale image classification benchmarks shows that our method achieves better performance than existing PTQ techniques at ultra-low precision levels of 2 to 4 bits.
Keyword (in Japanese) (See Japanese page) 
(in English) CNN / DNN / Post Training Quantization / Subset Quantization / Selective Two-word Logarithmic Quantization / / /  
Reference Info. IEICE Tech. Rep., vol. 124, no. 247, VLD2024-28, pp. 8-13, Nov. 2024.
Paper # VLD2024-28 
Date of Issue 2024-11-05 (VLD, ICD, DC, RECONF) 
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 VLD2024-28 ICD2024-46 DC2024-50 RECONF2024-58

Conference Information
Committee VLD DC RECONF ICD IPSJ-SLDM  
Conference Date 2024-11-12 - 2024-11-14 
Place (in Japanese) (See Japanese page) 
Place (in English) COMPAL HALL 
Topics (in Japanese) (See Japanese page) 
Topics (in English) Design Gaia 2024 -New Field of VLSI Design- 
Paper Information
Registration To VLD 
Conference Code 2024-11-VLD-DC-RECONF-ICD-SLDM 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Low-bit Quantization Methods for Neural Networks 
Sub Title (in English)  
Keyword(1) CNN  
Keyword(2) DNN  
Keyword(3) Post Training Quantization  
Keyword(4) Subset Quantization  
Keyword(5) Selective Two-word Logarithmic Quantization  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Emi Wada  
1st Author's Affiliation Waseda University (Waseda Univ.)
2nd Author's Name Shinji Kimuda  
2nd Author's Affiliation Waseda University (Waseda 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 2024-11-12 14:05:00 
Presentation Time 25 minutes 
Registration for VLD 
Paper # VLD2024-28, ICD2024-46, DC2024-50, RECONF2024-58 
Volume (vol) vol.124 
Number (no) no.247(VLD), no.248(ICD), no.249(DC), no.250(RECONF) 
Page pp.8-13 
#Pages 6 
Date of Issue 2024-11-05 (VLD, ICD, DC, RECONF) 

[Return to Top Page]

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