Information: Join today and make your research activities more affordable! Technical workshop participation fees and annual registration fees are available at member rates.
Notice: [Important] Announcement of Changes to Registration Fee Payment and Manuscript Upload Procedures for IEICE Technical Meetings
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 2024-09-27 14:15
Study on the effectiveness of the combination of QLoRA and Bayesian inference
Yusuke Takaya, Shozo Saeki, Takashi Sasaki, Minoru Kawahara, Hirohisa Aman (Ehime Univ.) NC2024-34
Abstract (in Japanese) (See Japanese page) 
(in English) In this study, we propose a novel solution called Laplace-QLoRA to address the critical issues of "high cost and overfitting" in fine-tuning large language models (LLMs). Specifically, Laplace-QLoRA tackles the problem of reducing costs and mitigating overfitting in the fine-tuning process. The proposed method first applies QLoRA to fine-tune a pre-trained model and then uses Laplace approximation to calculate the posterior distribution. This approach aims to improve model accuracy and optimize resource usage. Experimental results show that Laplace-QLoRA outperforms conventional fine-tuning methods in terms of memory efficiency and inference time. Notably, it also achieved high prediction accuracy on small datasets.
Keyword (in Japanese) (See Japanese page) 
(in English) Fine-Tune / Overfitting / QLoRA / Bayesian inference / Laplace approximation / / /  
Reference Info. IEICE Tech. Rep., vol. 124, no. 194, NC2024-34, pp. 11-16, Sept. 2024.
Paper # NC2024-34 
Date of Issue 2024-09-20 (NC) 
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 NC2024-34

Conference Information
Committee NC MBE  
Conference Date 2024-09-27 - 2024-09-28 
Place (in Japanese) (See Japanese page) 
Place (in English) Tohoku Univ. 
Topics (in Japanese) (See Japanese page) 
Topics (in English) NC, ME 
Paper Information
Registration To NC 
Conference Code 2024-09-NC-MBE 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Study on the effectiveness of the combination of QLoRA and Bayesian inference 
Sub Title (in English)  
Keyword(1) Fine-Tune  
Keyword(2) Overfitting  
Keyword(3) QLoRA  
Keyword(4) Bayesian inference  
Keyword(5) Laplace approximation  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Yusuke Takaya  
1st Author's Affiliation Ehime University (Ehime Univ.)
2nd Author's Name Shozo Saeki  
2nd Author's Affiliation Ehime University (Ehime Univ.)
3rd Author's Name Takashi Sasaki  
3rd Author's Affiliation Ehime University (Ehime Univ.)
4th Author's Name Minoru Kawahara  
4th Author's Affiliation Ehime University (Ehime Univ.)
5th Author's Name Hirohisa Aman  
5th Author's Affiliation Ehime University (Ehime Univ.)
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 2024-09-27 14:15:00 
Presentation Time 25 minutes 
Registration for NC 
Paper # NC2024-34 
Volume (vol) vol.124 
Number (no) no.194 
Page pp.11-16 
#Pages 6 
Date of Issue 2024-09-20 (NC) 

[Return to Top Page]

[Return to IEICE Web Page]
The Institute of Electronics, Information and Communication Engineers (IEICE), Japan