Abstract:
This research explores the application of FL to the MNIST dataset, focusing on impact of system heterogeneity. The study involves two clients collaboratively training a machine learning model using the FedAvg algorithm, with the aim of preserving data privacy and decentralization. We systematically analyze the behavior of the clients by varying the number of training epochs and evaluating the model’s accuracy under different data distributions. We investigate how the system heterogeneity impacts model convergence and performance consistency across clients. This work emphasizes the theoretical and algorithmic aspects of federated learning. The findings offer insights into the trade-offs between communication efficiency, accuracy, and data distribution in FL applications.
