Machine Learning Security and Privacy

Abstract

Our special issue explores emerging security and privacy aspects related to machine learning and artificial intelligence techniques, which are increasingly deployed for automated decisions in many critical applications today. With the advancement of machine learning and deep learning and their use in health care, finance, autonomous vehicles, personalized recommendations, and cybersecurity, understanding the security and privacy vulnerabilities of these methods and developing resilient defenses becomes extremely important. An area of research called adversarial machine learning has been developed at the intersection of cybersecurity and machine learning to understand the security of machine learning in various settings. Early work in adversarial machine learning showed the existence of adversarial examples: data samples that can create misclassifications at deployment time. Other threats against machine learning include poisoning attacks, where an adversary controls a subset of data at training time, and privacy attacks in which an adversary is interested in learning sensitive information about the training data and model parameters.