Privacy protection in machine learning: The state-of-the-art for a private decision tree

The explosive growth and widespread accessibility of digital data have led to a surge of research activity in the machine learning field. Typically a massive data collection is required to increase the quality of machine learning result. Often, these data contained highly sensitive information such as medical history, or financial records. Hence, privacy concerns have overshadowed by other factors in today's machine learning systems. A fundamental problem in privacy-preserving machine learning (PPML) is how to make the right tradeoff between privacy and utility. On the one hand, the PPML solution must not allow the original data records (e.g., training data) to be adequately recovered (i.e., privacy loss). On the other, it must allow the system to learn the model that is closely approximates to the model that is trained using the original data (i.e., utility gain). In this chapter, we will discuss several emerging technologies that can be used to protect privacy in machine learning systems. In addition, we also provide a state-of-the-art of the adoption of privacy preserving schemes in decision tree algorithms.