Decentralization Feature

[meshtiwa]

Based on the concept of Decentralized Federated Learning, use the Decision Tree and paper to implement using decentralization option (rather than centralized).

[royshadmon]

would be good to identify what the setup entails and provide an example on decentralized FL in a demo section + readme instruction.

[meshtiwa]

5-28:

• configure to differentiate between normal (CFL) vs DFL. Right now, it's hardcoded to be DFL
• test with 4 training nodes + different minParams
• (new todo to make later that spins off this) fully offload from aggregator. Right now, the aggregator is used as proxy/router to route the DFL to work with the training nodes. We ideally want to only send curl requests to training nodes, and not have the aggregator up AT ALL.