I'm new to Federated Learning so please bear with me. I'm working on a university project where I'd like to build a decentralized peer-to-peer network so that I can train models without data sharing and without the need for an aggregator (a central server with a joint model). The idea is to take TensorFlow Federated and add some client state and possibly use Gossip protocol to enable clients to talk to each other. I don't know if it's possible/feasible with TensorFlow Federated. I'd appreciate any feedback on this!
I tried looking for examples online but didn't find anything.