Differentially Private Bayesian Learning on Distributed Data
This work addresses privacy concerns in distributed machine learning applications, such as health care, by enabling differential privacy without requiring a single trusted party to access all data.
The paper tackles the problem of guaranteeing privacy in Bayesian learning when data is distributed across multiple parties, proposing a method that uses secure multi-party summation and the Gaussian mechanism to achieve differential privacy with asymptotically optimal inference and minimal extra computational cost.
Many applications of machine learning, for example in health care, would benefit from methods that can guarantee privacy of data subjects. Differential privacy (DP) has become established as a standard for protecting learning results. The standard DP algorithms require a single trusted party to have access to the entire data, which is a clear weakness. We consider DP Bayesian learning in a distributed setting, where each party only holds a single sample or a few samples of the data. We propose a learning strategy based on a secure multi-party sum function for aggregating summaries from data holders and the Gaussian mechanism for DP. Our method builds on an asymptotically optimal and practically efficient DP Bayesian inference with rapidly diminishing extra cost.