Privacy for Free: Wireless Federated Learning via Uncoded Transmission With Adaptive Power Control

Federated Learning (FL) refers to distributed protocols that avoid direct raw data exchange among the participating devices while training for a common learning task. This way, FL can potentially reduce information on local sets is leaked via communications. In order provide formal privacy guarantees, however, it generally necessary put in place additional masking mechanisms. When implemented wireless systems uncoded transmission, channel noise directly act as privacy-inducing mechanism. paper demonstrates that, long constraint level, measured differential (DP), below threshold decreases with signal-to-noise ratio (SNR), transmission achieves “for free”, i.e., without affecting performance. More generally, this work studies adaptive power allocation (PA) gradient descent aim of minimizing optimality gap under and constraints. Both orthogonal multiple access (OMA) non-orthogonal (NOMA) “over-the-air-computing” are studied, solutions obtained closed form an offline optimization setting. Furthermore, heuristic online methods proposed leverage iterative one-step-ahead optimization. The importance dynamic PA potential benefits NOMA versus OMA demonstrated through extensive simulations.