Differentially Private Mixture of Generative Neural Networks

TL;DR

This paper introduces a novel differentially private method for generating high-dimensional synthetic data using a mixture of neural networks, ensuring privacy while maintaining data utility.

Contribution

It proposes a new approach combining differentially private kernel k-means clustering with multiple generative neural networks for private data release.

Findings

Produces realistic synthetic samples from diverse datasets.

Accurately answers counting queries on synthetic data.

Maintains privacy with high data utility.

Abstract

Generative models are used in a wide range of applications building on large amounts of contextually rich information. Due to possible privacy violations of the individuals whose data is used to train these models, however, publishing or sharing generative models is not always viable. In this paper, we present a novel technique for privately releasing generative models and entire high-dimensional datasets produced by these models. We model the generator distribution of the training data with a mixture of $k$ generative neural networks. These are trained together and collectively learn the generator distribution of a dataset. Data is divided into $k$ clusters, using a novel differentially private kernel $k$-means, then each cluster is given to separate generative neural networks, such as Restricted Boltzmann Machines or Variational Autoencoders, which are trained only on their own cluster using differentially private gradient descent. We evaluate our approach using the MNIST dataset, as well as call detail records and transit datasets, showing that it produces realistic synthetic samples, which can also be used to accurately compute arbitrary number of counting queries.