Methods and Approaches for Privacy-Preserving Machine Learning

Abstract

One of the main problems of machine learning is the need for a large amount of memory and a long learning time. To solve this problem, many companies prefer to store their data and training models on remote servers. However, not all data and models can be stored in the plaintext without any protection. In many areas (for example, banking or medical), the privacy of data and models is very important. To ensure confidentiality, a privacy-preserving machine learning application is a good solution. This article discusses two main approaches to privacy-preserving machine learning (cryptographic and perturbation), describes methods for ensuring privacy, which they include, and provides examples of using of some methods in practice.