Scalable Differentially Private Data Generation via Private Aggregation of Teacher Ensembles
We present a novel approach named G-PATE for training differentially private data generator. The generator can be used to produce synthetic datasets with strong privacy guarantee while preserving high data utility. Our approach leverages generative adversarial nets to generate data and exploits the PATE (Private Aggregation of Teacher Ensembles) framework to protect data privacy. Compared to existing methods, our approach significantly improves the use of privacy budget. This is possible since we only need to ensure differential privacy for the generator, which is the part of the model that actually needs to be published for private data generation. To achieve this, we connect a student generator with an ensemble of teacher discriminators and propose a private gradient aggregation mechanism to ensure differential privacy on all the information that flows from the teacher discriminators to the student generator. Theoretically, we prove that our algorithm ensures differential privacy for the generator. Empirically, we provide thorough experiments to demonstrate the superiority of our method over prior work on both image and non-image datasets.
PDF Abstract
Fashion-MNIST
