Privacy Threats and Protection in Machine Learning
Jiliang Zhang, Chen Li, Jing Ye, Gang Qu
Abstract
With the improvement of computing power and storage level, Machine Learning (ML), especially Deep Learning (DL), has shown its capabilities beyond humans in areas such as image recognition, speech processing, and content recommendation. However, the data collected to build ML models often contains sensitive information, and models may have high commercial value. Compared with the security problem of model prediction errors caused by malicious external influences, privacy threats have not attracted widespread attention, and they have characteristics that are difficult to define and detect. This article reviews recent research progress on ML privacy. First, the privacy threats on data and models in different scenarios are described in detail. Then, typical privacy protection methods are introduced. Finally, the limitations and future development trends of ML privacy research are discussed.