Federated GAN

October 1, 2025 → Present

A distributed deep learning project that demonstrates how Generative Adversarial Networks (GANs) can be trained collaboratively across multiple clients without sharing raw data. The system simulates real-world heterogeneity by assigning different datasets and batch sizes to each client, while a central server coordinates training using the FedAvg aggregation algorithm. By combining privacy-preserving federated learning with GAN-based generation, the project showcases efficient, scalable training, robust handling of diverse computational capabilities, and comprehensive monitoring of model performance across communication rounds.