Federated De-confounding for Spatio-Temporal Forecasting
July 28, 2026
A novel federated learning framework treats diverse client heterogeneity as distinct causal environments rather than an optimization hurdle. By exploiting environmental diversity, the method improves generalization for spatio-temporal forecasting across shifting conditions.
HOW THIS AFFECTS YOU
●
researcherThis provides a new causal perspective on handling client heterogeneity in federated learning.