Decentralized Image Classification Using Federated Averaging Algorithm

The Decentralized Image Classification Using Federated Averaging (FedAvg) Algorithm is a privacy-preserving machine learning system that trains image classification models on users’ local devices without sharing actual image data. Traditional image classification systems collect and store all images on a central server, which poses serious privacy, security, and scalability concerns. This proposed system addresses these issues by implementing Federated Learning, where each participating device trains a local model using its own data and only shares model updates (weights) with the central server. The server then aggregates these updates using the Federated Averaging algorithm to construct a robust global model, which is redistributed back to client devices. This approach enables accurate and collaborative learning across distributed data sources without exposing raw image data. The system was built using Python and Streamlit, tested on the CIFAR-10 dataset across five simulated clients, and achieved high classification accuracy across ten object categories.