Paper Abstract and Keywords |
Presentation |
2019-03-06 10:30
Wider neural networks with ReLU activation generalize better Yasutaka Furusho, Kazushi Ikeda (NAIST) IBISML2018-111 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
Model size determination is important in machine learning since a larger model leads to overfitting, that is, a small training loss and a high test loss. Surprisingly, a multi-layer perceptron (MLP) with the ReLU activation function has a smaller training loss as well as a smaller gap between the training loss and a test loss as each layer gets wide.
To elucidate the mechanism, this paper theoretically derives an upper-bound of the gap for a random network and shows a wider MLP has a smaller gap.
Our numerical experiments confirmed the validity of our analysis and the applicability to the trained model. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
Neural networks / model size / over-fitting / generalization ability / / / / |
Reference Info. |
IEICE Tech. Rep., vol. 118, no. 472, IBISML2018-111, pp. 45-50, March 2019. |
Paper # |
IBISML2018-111 |
Date of Issue |
2019-02-26 (IBISML) |
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 |
IBISML2018-111 |
Conference Information |
Committee |
IBISML |
Conference Date |
2019-03-05 - 2019-03-06 |
Place (in Japanese) |
(See Japanese page) |
Place (in English) |
RIKEN AIP |
Topics (in Japanese) |
(See Japanese page) |
Topics (in English) |
Machine learning, etc. |
Paper Information |
Registration To |
IBISML |
Conference Code |
2019-03-IBISML |
Language |
Japanese |
Title (in Japanese) |
(See Japanese page) |
Sub Title (in Japanese) |
(See Japanese page) |
Title (in English) |
Wider neural networks with ReLU activation generalize better |
Sub Title (in English) |
|
Keyword(1) |
Neural networks |
Keyword(2) |
model size |
Keyword(3) |
over-fitting |
Keyword(4) |
generalization ability |
Keyword(5) |
|
Keyword(6) |
|
Keyword(7) |
|
Keyword(8) |
|
1st Author's Name |
Yasutaka Furusho |
1st Author's Affiliation |
Nara Institute of Science and Technology (NAIST) |
2nd Author's Name |
Kazushi Ikeda |
2nd Author's Affiliation |
Nara Institute of Science and Technology (NAIST) |
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 |
2019-03-06 10:30:00 |
Presentation Time |
30 minutes |
Registration for |
IBISML |
Paper # |
IBISML2018-111 |
Volume (vol) |
vol.118 |
Number (no) |
no.472 |
Page |
pp.45-50 |
#Pages |
6 |
Date of Issue |
2019-02-26 (IBISML) |