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 2017-11-07 16:00
[Invited Talk] Researchs on high-speed and efficient Deep Learning technologies
Takuya Fukagai, Koichi Shirahata, Yasumoto Tomita, Tetsutaro Hashimoto, Atsushi Ike, Masafumi Yamazaki, Akihiko Kasagi, Tsuguchika Tabaru (Fujitsu Lab. Ltd.), Liuan Wang, Song Wang, Li Sun, Jun Sun (FRDC) CPM2017-86 ICD2017-45 IE2017-71 Link to ES Tech. Rep. Archives: CPM2017-86 ICD2017-45
Abstract (in Japanese) (See Japanese page) 
(in English) Fujitsu laboratories have been doing Research and Development on AI technologies called "Human Centric AI Zinrai".

We have been advancing the Research and Development on Deep Learning Techonologies with the concept of "fast", "widely used" and "easy to use". We introduce the following researchs.

1. Memory reduction method for deep neural network training
2. Acceleration of a Deep Learning Framework with MPI
3. Column Weight Pruning for Accelerating DNN Inferences
4. Fast Algorithm Using Summed Area Tables with Unified Layer Performing Convolution and Average Pooling
5. An Automated CNN Recommendation System for Image Classification Tasks

1. is the research which improves the reusabilities of memories during the training phase of Deep Neural Networks. It enables us to reduce the GPU memory usage.

2. is the research which realizes the speed-up of the distributed training of Deep Neural Networks. This method utilizes MPI functions efficienlty.

3. is the research which realizes the fast recognition process of Deep Neural Networks. It reduces the amout of calculation and parameters of convolutional neural networks so that it can make use of the fast matrix calculation functionalities of the GPUs.

4. is the research about the algorithm which realizes the efficient calculation of a convolutional layer followed by an averagin-pooling layer. It performs the calculation of a pair formed by a convolutional layer and the following average-pooling layer without the computation of convolution.

5. is the research which makes it possible to suggest suitable Convolutional Neural Networks according to the given datasets. This method estimates the complexity score of the classification task as well as the classification ability score of the Convolutional Neural Networks.
Keyword (in Japanese) (See Japanese page) 
(in English) Deep Learning / Neural Network / GPU / CNN / Convolutional Neural Network / Convolution / MPI / algorithm  
Reference Info. IEICE Tech. Rep., vol. 117, no. 276, ICD2017-45, pp. 39-41, Nov. 2017.
Paper # ICD2017-45 
Date of Issue 2017-10-30 (CPM, ICD, IE) 
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 CPM2017-86 ICD2017-45 IE2017-71 Link to ES Tech. Rep. Archives: CPM2017-86 ICD2017-45

Conference Information
Committee VLD DC CPSY RECONF CPM ICD IE IPSJ-SLDM 
Conference Date 2017-11-06 - 2017-11-08 
Place (in Japanese) (See Japanese page) 
Place (in English) Kumamoto-Kenminkouryukan Parea 
Topics (in Japanese) (See Japanese page) 
Topics (in English) Design Gaia 2017 -New Field of VLSI Design- 
Paper Information
Registration To ICD 
Conference Code 2017-11-VLD-DC-CPSY-RECONF-CPM-ICD-IE-SLDM-EMB-ARC 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Researchs on high-speed and efficient Deep Learning technologies 
Sub Title (in English)  
Keyword(1) Deep Learning  
Keyword(2) Neural Network  
Keyword(3) GPU  
Keyword(4) CNN  
Keyword(5) Convolutional Neural Network  
Keyword(6) Convolution  
Keyword(7) MPI  
Keyword(8) algorithm  
1st Author's Name Takuya Fukagai  
1st Author's Affiliation Fujitsu Laboratories (Fujitsu Lab. Ltd.)
2nd Author's Name Koichi Shirahata  
2nd Author's Affiliation Fujitsu Laboratories (Fujitsu Lab. Ltd.)
3rd Author's Name Yasumoto Tomita  
3rd Author's Affiliation Fujitsu Laboratories (Fujitsu Lab. Ltd.)
4th Author's Name Tetsutaro Hashimoto  
4th Author's Affiliation Fujitsu Laboratories (Fujitsu Lab. Ltd.)
5th Author's Name Atsushi Ike  
5th Author's Affiliation Fujitsu Laboratories (Fujitsu Lab. Ltd.)
6th Author's Name Masafumi Yamazaki  
6th Author's Affiliation Fujitsu Laboratories (Fujitsu Lab. Ltd.)
7th Author's Name Akihiko Kasagi  
7th Author's Affiliation Fujitsu Laboratories (Fujitsu Lab. Ltd.)
8th Author's Name Tsuguchika Tabaru  
8th Author's Affiliation Fujitsu Laboratories (Fujitsu Lab. Ltd.)
9th Author's Name Liuan Wang  
9th Author's Affiliation Fujitsu Research and Development Center Co., Ltd. (FRDC)
10th Author's Name Song Wang  
10th Author's Affiliation Fujitsu Research and Development Center Co., Ltd. (FRDC)
11th Author's Name Li Sun  
11th Author's Affiliation Fujitsu Research and Development Center Co., Ltd. (FRDC)
12th Author's Name Jun Sun  
12th Author's Affiliation Fujitsu Research and Development Center Co., Ltd. (FRDC)
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 ()
Speaker Author-1 
Date Time 2017-11-07 16:00:00 
Presentation Time 45 minutes 
Registration for ICD 
Paper # CPM2017-86, ICD2017-45, IE2017-71 
Volume (vol) vol.117 
Number (no) no.275(CPM), no.276(ICD), no.277(IE) 
Page pp.39-41 
#Pages
Date of Issue 2017-10-30 (CPM, ICD, IE) 


[Return to Top Page]

[Return to IEICE Web Page]


The Institute of Electronics, Information and Communication Engineers (IEICE), Japan