AutoGAN-based Dimension Reduction for Privacy Preservation

Exploiting data and concurrently protecting sensitive information to whom data belongs is an emerging research area in data mining. Several methods have been introduced to protect individual privacy and at the same time maximize data utility. Unfortunately, existing techniques such as differential privacy are not effectively protecting data owner privacy in the scenarios using visualizable data (e.g., images, videos). Furthermore, such techniques usually result in low performance with a high number of queries. To address these problems, we propose a dimension reduction-based method for privacy preservation. This method generates dimensionally-reduced data for performing machine learning tasks and prevents a strong adversary from reconstructing the original data. In this paper, we first introduce a theoretical tool to evaluate dimension reduction-based privacy preserving mechanisms, then propose a non-linear dimension reduction framework using state-of-the-art neural network structures for privacy preservation. In the experiments, we test our method on popular face image datasets and show that our method can retain data utility and resist data reconstruction, thus protecting privacy.