Adversarial Machine Learning in Cybersecurity

Abstract

Adversarial machine learning algorithms deal with adversarial sample generation which is creating false input data that are capable enough to fool any machine learning model. For instance, attributes of a goodware can be added to a malware executable to make the classifier identify a malicious sample as benign. As the name suggests, “adversary” means opponent or enemy. If you are thinking what an enemy has got to do in machine learning, this chapter will take you through how vulnerable machine learning models are and how easily they can misunderstand during the learning process. If any set of input data when given to a machine learning model gets misclassified, we call them as adversarial samples.