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 2017-10-13 09:15
Improvement of speed using low precision arithmetic in deep learning and performance evaluation of accelerator
Hiroki Naganuma, Akira Sekiya, Kazuki Osawa, Hiroyuki Ootomo, Yuji Kuwamura, Rio Yokota (Tokyo Inst. of Tech.) PRMU2017-81
Abstract (in Japanese) (See Japanese page) 
(in English) While recent convolution neural networks (CNN)cite{ref:CNN} are improving performance, amout of computation and data volume are increasing.
In this research, by applying low precision to the learned model,
in addition to reduce data of CNN model, speeding up data access,
for a layer that is computation-bound, we propose a method to speed up by using a half precision floating point SIMD instruction.
We examined the influence of CNN recognition accuracy, the tendency of speeding up for each layer and its reason ,when we apply our method.
Furthermore, we conducted a performance evaluation for each accelerator (NVIDIAGTX1080TI, NVIDIA Pascal100 SXM2).
Keyword (in Japanese) (See Japanese page) 
(in English) image recognition / convolutional neural network / low-precision / half-precision / quantization / / /  
Reference Info. IEICE Tech. Rep., vol. 117, no. 238, PRMU2017-81, pp. 101-107, Oct. 2017.
Paper # PRMU2017-81 
Date of Issue 2017-10-05 (PRMU) 
ISSN Print edition: ISSN 0913-5685    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 PRMU2017-81

Conference Information
Committee PRMU  
Conference Date 2017-10-12 - 2017-10-13 
Place (in Japanese) (See Japanese page) 
Place (in English)  
Topics (in Japanese) (See Japanese page) 
Topics (in English)  
Paper Information
Registration To PRMU 
Conference Code 2017-10-PRMU 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Improvement of speed using low precision arithmetic in deep learning and performance evaluation of accelerator 
Sub Title (in English)  
Keyword(1) image recognition  
Keyword(2) convolutional neural network  
Keyword(3) low-precision  
Keyword(4) half-precision  
Keyword(5) quantization  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Hiroki Naganuma  
1st Author's Affiliation Tokyo Institute of Technology (Tokyo Inst. of Tech.)
2nd Author's Name Akira Sekiya  
2nd Author's Affiliation Tokyo Institute of Technology (Tokyo Inst. of Tech.)
3rd Author's Name Kazuki Osawa  
3rd Author's Affiliation Tokyo Institute of Technology (Tokyo Inst. of Tech.)
4th Author's Name Hiroyuki Ootomo  
4th Author's Affiliation Tokyo Institute of Technology (Tokyo Inst. of Tech.)
5th Author's Name Yuji Kuwamura  
5th Author's Affiliation Tokyo Institute of Technology (Tokyo Inst. of Tech.)
6th Author's Name Rio Yokota  
6th Author's Affiliation Tokyo Institute of Technology (Tokyo Inst. of Tech.)
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 2017-10-13 09:15:00 
Presentation Time 30 minutes 
Registration for PRMU 
Paper # PRMU2017-81 
Volume (vol) vol.117 
Number (no) no.238 
Page pp.101-107 
#Pages 7 
Date of Issue 2017-10-05 (PRMU) 

[Return to Top Page]

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