Best of Both Worlds in Federated LSA: Speedup When Possible, Personalization Always
PF-LSA is a new method that combines local agent updates with an average update across all agents. It achieves a balance between fast convergence (like federated averaging) and personalization (converging to individual agent solutions). The performance depends on the level of heterogeneity; it works best in low-heterogeneity settings for linear speedup.