PATE-TabTransGAN: Differentially Private Synthetic Tabular Data Generation via Transformer-Based Student Discrimination
Researchers have developed PATE-TabTransGAN, a novel framework for generating synthetic tabular data that adheres to formal differential privacy guarantees. This method combines the Private Aggregation of Teacher Ensembles (PATE) mechanism with a Transformer-based student discriminator to effectively model inter-feature dependencies while ensuring strong privacy. Experiments on four benchmark datasets demonstrated that PATE-TabTransGAN achieves competitive or superior performance in terms of AUROC and AUCPR compared to existing state-of-the-art differentially private synthesis methods. AI
IMPACT This research advances the state-of-the-art in privacy-preserving synthetic data generation, potentially enabling more secure use of sensitive tabular datasets.