dynamic-environment

FedCausal-Dyn is a novel federated learning framework that addresses dynamic feature drift by separating causal features from spurious variations, enabling robust prototype aggregation. It achieves state-of-the-art performance on federated domain generalization benchmarks.