FeGAN

What is it about?

FeGAN is a system to efficiently and robustly train generative adversarial networks (GANs) in the federated learning scheme. GAN is an impressive machine learning technique that allows the generation of data (that looks real) from noise. Training GANs over edge devices (via federated learning) is challenging due to the heterogeneity of devices' capabilities and the skewness of data distributions they have. We solve such problems with FeGAN, which we show achieves higher performance compared to existing solutions of training GANs.

Featured Image

Why is it important?

Generative adversarial networks (GANs) have many applications that are useful for many domains. Training GANs on edge devices (e.g., smartphones) is an open problem. FeGAN efficiently and robustly solves this problem, showing great performance and scalability.