Privacy-preserving machine learning based on secure two-party computations

Abstract

The paper is devoted to the analysis of privacy-preserving machine learning systems based on secure two-party computations. The paper provides introductory information about privacy-preserving machine learning systems, analyses the goals and objectives of its application. A generalized model of privacy-preserving machine learning architecture is proposed, reflecting the main functional blocks of the systems. The formulation of the problem of secure multi-party computation is considered. The descriptions of cryptographic primitives and protocols used to implement two-party secure computation protocols, including homomorphic encryption, secret sharing schemes, oblivious transfer and garbled circuits, are given. The current privacy-preserving machine learning systems based on two-party secure computations are analyzed. The main attention is paid to algorithmic aspects of systems, methods and protocols of information security in them. Systems resistant to semi-honest and active adversaries are considered, both based on universal modules for secure two-party computations, and specialized ones designed to ensure the privacy of specific machine learning technologies, such as convolutional neural networks. Implemented prototypes of several such systems are considered in detail. Based on the results of the analysis, conclusions are formulated about the features of the future privacy-preserving machine learning systems.

Keywords

• privacy-preserving machine learning, secure multi-party computations, garbled circuits, oblivious transfer, secret sharing scheme, homomorphic encryption.