FedCausal-Dyn: A Causal-Dynamic Paradigm for Federated Learning under Dynamic Feature Drift

This paper addresses the challenging problem of dynamic feature drift in federated learning, where data distributions evolve across clients and over time – a common scenario in real-world applications like financial technology. Existing approaches often assume static drift, limiting their effectiveness in non-stationary environments. To overcome this, we propose $\mathbf{\text{FedCausal-Dyn}}$, a novel federated learning framework built on a causal-dynamic paradigm. Its key innovation is $\mathit{\text{causal-domain}}\mathit{\text{feature}}\mathit{\text{separation}}$, which disentangles domain-invariant causal features from spurious, domain-specific variations via specialized projection heads and adversarial training. This enables $\mathit{\text{reliable}}\mathit{\text{and}}\mathit{\text{dynamic}}\mathit{\text{prototype}}\mathit{\text{aggregation}}$, weighting local class prototypes by estimated reliability before global aggregation. We further introduce $\mathit{\text{causal-feature}}\mathit{\text{guided}}\mathit{\text{collaborative}}\mathit{\text{regularization}}$, unifying prototype contrastive alignment and domain invariance into a cohesive objective. Extensive experiments on three federated domain generalization benchmarks demonstrate that FedCausal-Dyn consistently achieves state-of-the-art performance, with the highest average accuracy and the most stable results. Ablation studies confirm each component’s critical contribution. Our work provides a robust and principled solution for federated learning under dynamic feature drift.