Awesome

Federated-Learning

We implement the paper Communication-Efficient Learning of Deep Networks from Decentralized Data. Its blog is here.

We wish to explore several possible approaches:

1. server pulling gradient update from clients;
2. clients eagerly train and push updates to server;
3. extend to multi-server; servers gossip to sync;
4. client-only p2p gradient updates