Abstract: Bayesian methods are frequently used for analysing privacy-sensitive datasets, including medical records, emails, and educational data, and there is a growing need for practical Bayesian inference algorithms that protect the privacy of individuals' data. To this
end, we provide a general framework for privacy-preserving variational Bayes (VB) for a large class of probabilistic models, called the conjugate exponential (CE) family. Our primary observation is that when models are in the CE family, we can privatise the variational posterior distributions simply by perturbing the expected sufficient statistics of the complete-data likelihood. For widely used non-CE models with binomial likelihoods (e.g., logistic regression), we exploit the Polya-Gamma data augmentation scheme to bring such models into the CE family, such that inferences in the modified model resemble the original (private) variational Bayes algorithm as closely as possible. The iterative nature of variational Bayes presents a further challenge for privacy preservation, as each iteration increases the amount of noise needed. We overcome this challenge by combining: (1) a relaxed notion of differential privacy, called concentrated differential privacy, which provides a tight bound on the privacy cost of multiple VB iterations and thus significantly decreases the amount of additive noise; and (2) the privacy amplification effect of subsampling mini-batches from large-scale data in stochastic learning. We empirically demonstrate the effectiveness of our method in CE and non-CE models including latent
Dirichlet allocation (LDA), Bayesian logistic regression, and Sigmoid Belief Networks
(SBNs), evaluated on real-world datasets. |