A Deep Learning-based Data Usability Enhancement Scheme for Differential Privacy

While differential privacy (DP) is widely used to ensure privacy, it can also significantly reduce data accuracy. Current research attempts to improve accuracy by leveraging post-processing techniques, but these methods are sub-optimal and only applicable to specific data types. To address the issue, in this work, we propose a novel deep learning-based data usability enhancement method for differential privacy that is data-type independent. By using image denoising technology, the proposed scheme reduces the mean square error (MSE) of DP-perturbed data and theoretically justifies the relationship between data noise reduction and image noise reduction. The effectiveness of the proposed scheme is demonstrated through a comprehensive evaluation.