Privacy Tradeoffs in Vertical Federated Learning

Linh Tran, Timothy Castiglia, Stacy Patterson, Ana Milanova

Keywords: Differential Privacy, Quantization, Multi-Party Computation

TL;DR: We present VFL-PBM, a communication-efficient Vertical Federated Learning algorithm with end-to-end Differential Privacy guarantees.

Abstract: We present VFL-PBM, a communication-efficient Vertical Federated Learning algorithm with Differential Privacy guarantees. VFL-PBM combines Secure Multi-Party Computation with the recently introduced Poisson Binomial Mechanism to protect parties' private datasets during model training. We analyze the end-to-end privacy and convergence behavior of our algorithm, and we provide the first theoretical characterization of the relationship between privacy, convergence error, and communication cost in differentially-private VFL. Our experiments show the VFL model performs well, with negligible decline in accuracy as we increase the privacy parameters.