Resumen
This article, written for the Robust Machine Learning Curriculum, discusses the so-called Generative Models in Machine Learning. Generative models learn the distribution of data from some sample data set and then can generate (create) new data instances. Generative models are popular tools with a wide range of applications. Recent advances in deep learning have led to improvements in the architecture of generative models, and some current models can (in some cases) produce realistic enough results to fool both end-users (humans) and recognition and classification algorithms. Generative models are used in constructing adversarial attacks. Instead of looking for minimal modifications, as in classical evasion attacks, generator models allow, for example, to create adversarial examples completely from scratch. At the same time, generator models are just as vulnerable to adversarial attacks as classifiers. This is our first material on this topic and, obviously, consideration of this important topic will be continued.