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 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
Date of Issue 2022-12-15 (IBISML) 


[Return to Top Page]

[Return to IEICE Web Page]


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