Kang Li is a professor of computer science and the director of the Institute for Cybersecurity and Privacy at the University of Georgia. His research results have been published at academic venues, such as IEEE S&P;, ACM CCS and NDSS, as well as industrial conferences, such as BlackHat, SyScan, and ShmooCon. Dr. Kang Li is the founder and mentor of multiple CTF security teams, including SecDawg and Blue-Lotus. He was a founder and player of Team Disekt, one of the finalist teams in the 2016 DARPA Cyber Grand Challenge. .
[Abstract]
==========
Advance in deep learning algorithms overshadows their security risk in software implementations. This presentation discloses a set of vulnerabilities in popular deep learning frameworks including TensorFlow, Torch and Caffe. Contrast to the small code size of deep learning models, these deep learning frameworks are complex and contains heavy dependencies on numerous open source packages. By exploiting these framework implementations, this presentation demonstrates attacks on common deep learning applications such as as voice recognition and imaging classifications. The cases to be demonstrated include denial-of-service attacks that crash or hang an deep learning applications, and control-flow hijacking attacks that cause either system compromise or recognition evasions.