FedPCD: Priority-Constrained Federated Aggregation

Name: FedPCD: Priority-Constrained Federated Aggregation
Uploaded: 2026-10-19T01:53:01.477638+00:00
Duration: 458 s
Description: FedPCD introduces a method to improve fairness in federated learning by balancing global model utility with individual client performance needs. It uses Priority-Constrained Descent (PCD) to keep the standard FedAvg update as primary while making only minimal necessary corrections to satisfy specific client progress constraints, leading to superior accuracy and faster convergence.

FedPCD introduces a method to improve fairness in federated learning by balancing global model utility with individual client performance needs. It uses Priority-Constrained Descent (PCD) to keep the standard FedAvg update as primary while making only minimal necessary corrections to satisfy specific client progress constraints, leading to superior accuracy and faster convergence.

Open the full episode