Privacy Attacks and Defenses in Machine Learning: A Survey

As machine learning has gradually become an important technology in the field of artificial intelligence, its development is also facing challenges in terms of privacy. This article aims to summarize the attack methods and defense strategies for machine learning models in recent years. Attack methods include embedding inversion attack, attribute inference attack, membership inference attack and model extraction attack, etc. Defense measures include but are not limited to homomorphic encryption, adversarial training, differential privacy, secure multi-party computation, etc., focusing on the analysis of privacy protection issues in machine learning, and providing certain references and references for related research.