Information Security Meets Adversarial Examples

Matthias Kirchner, Cecilia Pasquini, Ilia Shumailov

Abstract: Machine learning (ML) plays a pivotal role in supporting a variety of research problems at the core of information security and forensics (IFS). ML-based systems have improved overall performance in many applications, replacing more and more deterministic rule-based approaches with data-driven ones. Yet, enforcing their dependability in typical IFS scenarios that include strategic adversaries is a challenging open problem.This special session aims at strengthening the link between concepts and methods developed in the adversarial ML and IFS domains. To this end, it gathers contributions exploring implications of and countermeasures to adversarial attacks against learning systems. Special emphasis is placed on media security applications.