FedRouter introduces task-centric clustering approach to improve federated learning performance across diverse distributed datasets

FedRouter addresses limitations of personalized federated learning by building specialized models for individual tasks rather than individual clients

The approach tackles two key challenges: generalization when clients encounter unseen tasks or data distribution shifts, and interference between multiple data distributions within a single client

Uses adapter-based personalization to improve robustness and performance in heterogeneous federated learning environments

Clustering-based methodology enables better model aggregation across diverse task distributions in distributed and privacy-preserving training scenarios