Date: 13th Oct – SundayTime: 09:30-10:00Location: Ballroom A
Artificial intelligence is steadily flooding our world and slipping into a lot of verticals from autonomous cars and robots to defense, media, and smart homes. Hundreds of new startups implement AI solutions worldwide, and we are getting closer to the point where machine learning based solutions consume traditional algorithms. While we understand more or less how to deal with software vulnerabilities, we have no clue what’s happening in ML-based solutions. In addition, we don’t know how it’s possible to hack them. However, we cannot but take note of adversarial examples, which have recently attracted media attention but were invented 5 years ago. This area is rapidly growing and soon there will be almost 1000 research papers on this topic. Some of them will hit media only in 5 years or so.
In my presentation, I will show what occurs in AI security industry, which is the most closed cybersecurity area. I will address the most critical AI applications such as face recognition, self-driving cars, voice assistants, and the latest attacks. Then I will present ML algorithms such as classification, regression, reinforcement learning, clustering, etc. I will explain how to attack them. Finally, particular attack methods will be discussed such as adversarial, privacy, poisoning, backdoor, reprogramming, and you will see how they are evolving.