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 2022-01-20 15:00
[Short Paper] Evaluation of Few Round Training with Distillation-Based Semi-Supervised Federated Learning
Yuki Yoshida (Tokyo Tech), Sohei Itahara (Kyoto Univ.), Takayuki Nishio (Tokyo Tech) SeMI2021-65
Abstract (in Japanese) (See Japanese page) 
(in English) This paper studies how to reduce the number of rounds in model training using Distillation-based Semi-supervised federated learning (DS-FL). Federated Learning (FL) is a machine learning framework that trains a model with data stored in devices without sharing the data by sharing parameters of the updated model. In DS-FL, the output of the model, logit, is used instead of the model, which significantly reduces the communication traffic while achieving comparable accuracy to the existing FL. In this study, we focus on the logit aggregation of DS-FL and experimentally show the possibility of significantly reducing the sharing frequency of logit in DS-FL. Specifically, by increasing the number of model updates at each device, which is usually set to 5-10 times, the models at each device are over-fitted to their own data to generate a model that can make predictions with high accuracy for specific data. This improves the accuracy of the aggregated logit from the initial stage of training and achieves high prediction accuracy even with a small number of training rounds. Through machine learning experiments using an image classification task, we have shown that DS-FL can train models with the same level of accuracy as conventional models in less than five rounds for relatively simple tasks.
Keyword (in Japanese) (See Japanese page) 
(in English) Federated learning / Knowledge distillation / non-IID data / communication efficiency / Semi-supervised learning / / /  
Reference Info. IEICE Tech. Rep., vol. 121, no. 333, SeMI2021-65, pp. 48-50, Jan. 2022.
Paper # SeMI2021-65 
Date of Issue 2022-01-13 (SeMI) 
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 SeMI2021-65

Conference Information
Committee SeMI  
Conference Date 2022-01-20 - 2022-01-21 
Place (in Japanese) (See Japanese page) 
Place (in English)  
Topics (in Japanese) (See Japanese page) 
Topics (in English)  
Paper Information
Registration To SeMI 
Conference Code 2022-01-SeMI 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Evaluation of Few Round Training with Distillation-Based Semi-Supervised Federated Learning 
Sub Title (in English)  
Keyword(1) Federated learning  
Keyword(2) Knowledge distillation  
Keyword(3) non-IID data  
Keyword(4) communication efficiency  
Keyword(5) Semi-supervised learning  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Yuki Yoshida  
1st Author's Affiliation Tokyo Institute of Technology (Tokyo Tech)
2nd Author's Name Sohei Itahara  
2nd Author's Affiliation Kyoto University (Kyoto Univ.)
3rd Author's Name Takayuki Nishio  
3rd Author's Affiliation Tokyo Institute of Technology (Tokyo Tech)
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 2022-01-20 15:00:00 
Presentation Time 10 minutes 
Registration for SeMI 
Paper # SeMI2021-65 
Volume (vol) vol.121 
Number (no) no.333 
Page pp.48-50 
#Pages 3 
Date of Issue 2022-01-13 (SeMI) 

[Return to Top Page]

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