FDA-Opt: Federated Fine-Tuning via Dynamic Update Schedules

Name: FDA-Opt: Federated Fine-Tuning via Dynamic Update Schedules
Uploaded: 2026-08-20T13:22:55.856746+00:00
Duration: 752 s

The episode discusses the paper 'FDA-Opt: Federated Fine-Tuning via Dynamic Update Schedules,' which addresses communication bottlenecks in federated learning. The authors introduce Fda-Opt, an algorithm that dynamically determines when to stop local training based on model variance. This method is highly communication-efficient, more stable than existing methods like FedOpt, and requires no new hyperparameters.

Open the full episode