Differentially-Private Multi-Tier Federated Learning

While federated learning (FL) eliminates the transmission of raw data over a network, it is still vulnerable to privacy breaches from the communicated model parameters. In this work, we propose Multi-Tier Federated Learning with Multi-Tier Differential Privacy (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{M}^{2} \text{FDP}$</tex>), a DP-enhanced FL methodology for jointly optimizing privacy and performance in hierarchical networks. One of the key concepts of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$M^{2} F D P$</tex> is to adapt DP noise injection according to different tiers of an established edge/fog hierarchy (e.g., edge devices, intermediate nodes, and other layers up to cloud servers) according to the trust models within particular subnetworks. We conduct a comprehensive analysis of the convergence behavior of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$M^{2} F D P$</tex>, revealing conditions on parameter tuning under which the training process converges sublinearly to a finite stationarity gap that depends on the network hierarchy, trust model, and target privacy level. Subsequent numerical evaluations demonstrate that <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$M^{2} F D P$</tex> obtains substantial improvements in these metrics over baselines for different privacy budgets, and validate the impact of different system configurations.