Paper Abstract and Keywords |
Presentation |
2021-06-24 13:25
A Study of Ensemble Learning for Randomly Weighted Neural Network Yasuyuki Okoshi, Kazutoshi Hirose, Kota Ando, Kazushi Kawamura, Thiem Van Chu, Masato Motomura, Jaehoon Yu (Tokyo Tech) SIS2021-7 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
Recent research on deep learning shows the possibility of building neural networks by learning connection existences instead of weights, in which weights are random numbers and do not change. Connection-based learning can significantly reduce memory resources to store the huge number of weights. In this research, we adopt various ensemble learning methods to the randomly weighted neural networks for improving the trade-off between computational cost and accuracy. On CIFAR-100, our ensembled ResNet18 model achieved 3.0% higher accuracy than the original randomly weighted ResNet18 model. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
machine learning / deep learning / ensemble learning / neural network / image classification / / / |
Reference Info. |
IEICE Tech. Rep., vol. 121, no. 73, SIS2021-7, pp. 37-42, June 2021. |
Paper # |
SIS2021-7 |
Date of Issue |
2021-06-17 (SIS) |
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 |
SIS2021-7 |
|