When the Internet was all shiny and new, many rushed to deploy technology, applications, e-commerce and customer portals with only a fleeting concern for security. Fast forward a couple of decades and security has become one of the most important and pressing issues in the on-line world.
Machine Learning is currently sitting in that same shiny and new phase. Many organizations are racing to deploy their machine learning applications such as: policing using facial recognition, self-driving cars, to relying on machine learning algorithms to make critical business decisions, without giving careful consideration to potential security issues.
The first part of this presentation examines the security vulnerabilities of machine learning systems from algorithm design to training methodologies to operational deployment. You will learn about the types of attacks that could be used to subvert, compromise or render machine learning systems inoperable. We will cover the risk implications and look at how mission critical these systems can be without proper testing and security implementation.
The second part of this presentation describes how many of the security protocols, testing methods and best practices in other areas can be adapted to develop robust security models for machine learning systems.