Adversarial examples - vulnerability of machine learning methods and prevention

Datum konání: 06.04.2018
Přednášející: Petra Vidnerová
Odpovědná osoba: Kotera

Adversarial examples are examples specially designed to be misclassified by a trained machine learning model. We will show several ways how to generate such examples, including adversarial examples generated by genetic algorithm. We will demonstrate how various machine learning methods are vulnerable to adversarial examples. Finally, attempts to prevent adversarial examples will be discussed.