Machine Learning for Privacy Preservation with Minimal Information Leakage

A wide range of application sectors are progressively using machine learning (ML). A successful machine learning (ML) model often requires a lot of training data and powerful processing resources, especially the newly developing deep neural network model. Due to the possibility of information leakage and the changing legislative landscapes that progressively limit Privacy-sensitive data access and use, the need for large amounts of available data, raises significant privacy concerns. Additionally, adversarial techniques. Therefore, carefully thought-out privacy-preserving machine learning (PPML) solutions are essential and are generating more and more research attention from academia and business. More and more PPML initiatives are being put forth through the incorporation of privacy-preserving methods into machine learning algorithms. As existing PPML techniques intersect with ML, systems, security, and privacy, among others, there is an important requirement to understand current research, associated barriers, and roadmaps for future research. This study offers a PGU model to help evaluate various PPML solutions by intricately breaking down each solution's privacy-preserving functions. It also evaluates and summarizes existing privacy-preserving methodologies in a methodical manner. Phase, Guarantee, and Technical Utility make up the PGU model's trinity. We also address the special features and difficulties of PPML and suggest future study avenues that will be beneficial to a wide variety of research communities in the fields of Machine learning, security systems, distributed systems and privacy system.