Private Coupled Descent: dimension-free differential privacy for vertical federated learning by privatizing the small shared coupling variable, with convergence guarantees, matching bounds, and synthetic validation.

COVERT: private, communication-efficient vertical federated block-term tensor regression on clinical multi-view data (imaging, ECG, labs), at parity with centralized while transmitting only a DP-noised coupling score.