Adversarial machine learning (AL) is a relatively new and extremely active research domain. It is focused on understanding the susceptibility of machine learning algorithms to specially crafted inputs that are designed to mislead the learning algorithms into a wrong conclusion, while being perceptually indistinguishable from valid input. A series of discoveries in recent years in the AL domain have allowed practical attacks against real world systems, and as a result became one of the main issues concerning the AI community today.
This course is a journey through the evolution of adversarial machine learning in recent years. It starts with the early methods of attack and defense, and concludes with recent discoveries and outstanding research questions. As part of this journey we will review notable studies, and discuss their contribution to the understanding of this phenomenon. We will discuss the risks imposed by adversarial attacks to real life systems and finally address the domain’s main open question –
What is it that makes adversarial examples so difficult to defend against?The course is designed for
middle to senior level professionals, who have a **solid background in Deep Learning **and related concepts and have experience with DL tools and practices
The course prerequisites are:
- Robust knowledge of Python stack for DS and DL
- General understanding of generic neural net architectures.