An Analysis of Privacy Attacks in Supervised Machine Learning Algorithms

With the widespread integration of machine learning (ML) into diverse applications, it has become paramount to scrutinize its implications for security and privacy. Despite the burgeoning body of work in privacy, there is a noticeable gap in research attention between the security and privacy aspects of ML, specifically supervised machine learning (S-ML). This paper bridges this gap by analyzing research papers focusing on privacy attacks against S-ML over the past years. This study offers a comprehensive overview of prevalent defense mechanisms and initiates a thoughtful discussion on unresolved challenges and prospective research directions. As ML continues to permeate various domains, the scientific community's interest has expanded beyond performance metrics to encompass broader concerns such as security, privacy, fairness, and explainability. This survey investigates state-of-the-art privacy-related attacks, elucidating common design patterns and distinctive characteristics. Through this exploration, we identify several open problems that demand further investigation, emphasizing the critical need to fortify privacy frameworks alongside the escalating ubiquity of ML technologies.