Event
ECE Colloquium - Tudor Dumitras, UMD
Friday, September 26, 2025
3:30 p.m.-4:30 p.m.
Jeong H. Kim Engineering Building, Room 1110
Darcy Long
301 405 3114
dlong123@umd.edu
Speaker: Tudor Dumitras, Associate Professor, Department of Electrical and Computer Engineering, University of Maryland
Title: AI (in)Security or How I Learned to Love the Arms Race
Abstract: When we apply artificial intelligence to security tasks (e.g. malware detection), we often adopt off-the-shelf techniques that were developed for other domains, which have their own unique constraints. When we study the security of artificial intelligence, we often focus on narrow and artificial threat models (e.g. adversarial examples) that may not capture the capabilities and limitations of real adversaries. Moreover, the two fields seem radically at odds with each other, because AI is not designed to operate under continuous adversarial pressure, which is inevitable in the security domain.
In this talk I will draw on examples from both fields to argue that this tension also provides an opportunity for making progress. Anti-virus tools used to combat malware in the real world have implemented machine learning techniques for over a decade, while facing the adversarial pressure that occurs naturally in this setting. We can learn from this experience to rethink the threat models for AI-based systems. Our results show how realistic adversaries may succeed despite limited capabilities (e.g., through clean-label data poisoning), utilize new attack vectors (e.g. hardware-based), or pursue under-explored goals (e.g. slowdown). Such security-inspired insights are changing our understanding of AI risks, have important practical implications in several domains, and may yield a fresh perspective on the generalization gap in machine learning.
Bio: Tudor Dumitras is an Associate Professor in the Electrical & Computer Engineering Department at the University of Maryland, and director of the Bad Data Lab. His research is at the intersection of computer security and machine learning. His work was featured in the Research Highlights of the Communications of the ACM, and it was recognized with first place in the CSAW Applied Research Competition, a DARPA Young Faculty Award, and an honorable mention in the NSA competition for the Best Scientific Cybersecurity Paper. Several of his papers have received coverage in the popular press, including the MIT Technology Review, Forbes, the Cyberwire, and the Economist. He holds a Ph.D. degree from Carnegie Mellon University.
