FLArch: A Federated Learning Architecture Supporting Adding Different Modules

FLArch is a experimental framework for FL researchers to quickly experiment on FL.

Client

• Weight update pruning: support model pruning for weight update to accelerate the network transmission.
• Data poisoning attack: support simulation of data poisoning attack in clients
• Model poisoning attack: support simulation of model poinsoning attack in clients
• Personalization of local model: support local model personalization, optimization

Server

• Participant selection: support adding different participant selection modules
• Aggregation algorithm: support adding different aggregation algorithms at server
• Updates history preserve: support preserve the history for each model update
• Byzantine client detection: support Byzantine attacker detection process

Profile

• Network latency simulation: support simulate the network latency to profile network performance
• Reputation management: support reputation management for different clients