A Data Augmentation Method Based on Generative Adversarial Networks in Differential Privacy

In view of the failure of the data collector to receive the expected amount of data in the differential privacy scenario, we propose a data augmentation method based on generative adversarial model in differential privacy, which is implemented by adding noise to the generative adversarial network gradient. We use optimization strategies in the model to ensure that the model can converge. The experimental results show that the proposed method can generate data satisfying differential privacy under the condition of ensuring the convergence of the model.