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 |
3 |
Date of Issue |
2017-10-30 (CPM, ICD, IE) |
|