Paper Abstract and Keywords |
Presentation |
2022-12-23 09:40
Effect of memory unit initialization on performance for function approximation Yuto Terasawa, Jun Ohkubo (Saitama Univ.) IBISML2022-52 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
Many researchers have proposed various neural network models for learning time-series data, such as RNN, LSTM, and Transformer. Recently, a new method called `memory unit' has been proposed. In conventional RNNs and LSTMs, hidden layers and storage cells are black box elements. On the other hand, the memory unit uses explicit approximation using orthogonal polynomials. Moreover, the properties of orthogonal polynomials enable us to avoid learning the weights for the memory part. However, there remain some unclear points regarding the basic properties of the memory unit implementation. For example, the approximation accuracy of the memory unit decreases in first several steps. In this study, we perform several numerical experiments to improve the accuracy of function approximation of the memory unit. As a result, we confirmed that a preprocessing method based on the initial values of the input data improves the accuracy of function approximation. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
Neural Network / time series data / RNN / LSTM / memory units / / / |
Reference Info. |
IEICE Tech. Rep., vol. 122, no. 325, IBISML2022-52, pp. 62-69, Dec. 2022. |
Paper # |
IBISML2022-52 |
Date of Issue |
2022-12-15 (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 |
IBISML2022-52 |
Conference Information |
Committee |
IBISML |
Conference Date |
2022-12-22 - 2022-12-23 |
Place (in Japanese) |
(See Japanese page) |
Place (in English) |
Kyoto University |
Topics (in Japanese) |
(See Japanese page) |
Topics (in English) |
Machine Learning, etc. |
Paper Information |
Registration To |
IBISML |
Conference Code |
2022-12-IBISML |
Language |
Japanese |
Title (in Japanese) |
(See Japanese page) |
Sub Title (in Japanese) |
(See Japanese page) |
Title (in English) |
Effect of memory unit initialization on performance for function approximation |
Sub Title (in English) |
|
Keyword(1) |
Neural Network |
Keyword(2) |
time series data |
Keyword(3) |
RNN |
Keyword(4) |
LSTM |
Keyword(5) |
memory units |
Keyword(6) |
|
Keyword(7) |
|
Keyword(8) |
|
1st Author's Name |
Yuto Terasawa |
1st Author's Affiliation |
Saitama University (Saitama Univ.) |
2nd Author's Name |
Jun Ohkubo |
2nd Author's Affiliation |
Saitama University (Saitama Univ.) |
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 |
() |
Speaker |
Author-1 |
Date Time |
2022-12-23 09:40:00 |
Presentation Time |
20 minutes |
Registration for |
IBISML |
Paper # |
IBISML2022-52 |
Volume (vol) |
vol.122 |
Number (no) |
no.325 |
Page |
pp.62-69 |
#Pages |
8 |
Date of Issue |
2022-12-15 (IBISML) |
|