Organizer
Partner Company
Machine learning implementation security in the wild
Denis Kolegov, Anton Nikolaev
BiZone
Abstract
In this talk, we will present the results of the Internet-wide survey on implementation security of practical machine learning systems. We will show that many machine learning related systems suffer from low-hanging fruit implementation vulnerabilities that can compromise the security of machine learning. In our presentation, we will describe the methodology of the survey and the used automation framework. We will also disclose the vulnerabilities found in widespread and most popular machine learning products and technologies.
Speaker Info
Denis Kolegov is a principal security researcher at BiZone LLC and an associate professor of Computer Security at Tomsk State University.
His research focuses on network security, machine learning security, web application security, cryptography engineering and covert communications. He holds a PhD and an associate professor degree. Denis presented at various international security conferences including Power of Community, DeepSec, Area41, SecurityFest, Zero Nights, Positive Hack Days, InsomniHack and SibeCrypt.
Anton Nikolaev is a security developer at BiZone LLC and a post-graduate student at Computer Security department of Tomsk State University. Anton also is the lead developer of the open-source Grinder framework and a contributor of SD-WAN New Hope and AISec projects. He gave talks at different international security conferences, such as Zero Nights and Positive Hack Days.