Earticle

초록

영어

In this research, we have studied a federated learning-based image classification algorithm. In this regard, we have used Resnet18 classifier as a backbone deep learning algorithm for image classification. The training process of this algorithm involved federated learning, such as training is performed at multiple clients (Jetson Nano and TX2) and is averaged at the server (Jetson Xavier). We have exploited the Nvidia Docker container image to deploy our algorithms for the training process. For our experiments, we have used only two clients and one server during the training process of the Resnet18 image classifier. The extension of the CIFAR10 (50K to 500K samples) dataset has been used for training, known as EC10, which contained 1000 subsets for IID client distribution. We have validated the accuracy for both using the federated learning process and traditional training at several strategies and have presented the results correspondingly.

목차

Abstract
1. Introduction
2. Related Works
3. Methodology
4. Experiments
4.1. Experimental setup
4.2. Experimental result
5. Conclusions
Acknowledgment
References

키워드

저자

• 울라 샨 [ Shan Ullah | Department of Electrical and Computer Engineering, Inha University, Incheon, South Korea ]
• 키크마툴로 [ Khikmatullo Tulkinbekov | Department of Electrical and Computer Engineering, Inha University, Incheon, South Korea ]
• 김덕환 [ Deok-Hwan Kim | Department of Electrical and Computer Engineering, Inha University, Incheon, South Korea ] Corresponding Author

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  한국차세대컴퓨팅학회 학술대회

• 간기

  반년간

• 수록기간

  2021~2025

• 십진분류

  KDC 566 DDC 004