Secure and Efficient Image Classification using Decentralized Federated Learning with ResNet

Abstract

This research investigates a decentralized federated learning (DFL) strategy for image classification using the ResNet architecture. The primary objective is to improve the accuracy and privacy of image classification models by training them collaboratively across multiple decentralized clients without directly sharing sensitive data. The study focuses on enhancing model efficiency and robustness in a decentralized setting. Our results demonstrate the effectiveness of the proposed DFL strategy compared to centralized approaches, highlighting improvements in accuracy and robustness while preserving data privacy.

Introduction

Federated learning (FL) enables collaborative model training on decentralized data without compromising privacy. However, traditional FL often relies on a central server, creating a single point of failure and potential privacy risks. Decentralized FL addresses these issues by removing the central server, distributing coordination among participating clients. This is particularly relevant for sensitive data like medical images or personal photos where privacy is paramount. Existing DFL methods often struggle with efficient communication and maintaining model accuracy across diverse data distributions. This research aims to address these challenges by leveraging the robust performance of ResNet architecture within a novel DFL framework.

Key Features

• Decentralized federated learning without reliance on a central coordinator
• Utilization of ResNet deep convolutional networks for robust image feature extraction
• Privacy-preserving collaborative model training across multiple clients
• Efficient communication protocols minimizing bandwidth and latency overhead
• Robustness against heterogeneous data distributions across clients