Yigitcan Kaya
I am a postdoctoral fellow at UCSB’s SecLab, working with Giovanni Vigna and Chris Kruegel. I am supported by the Intelligence Community (IC) Postdoctoral Research Fellowship, awarded for my proposal on building AI-driven antifragile cyber defenses. I earned my PhD in Computer Science from the University of Maryland, College Park, advised by Tudor Dumitraş.
I work at the intersection of trustworthy AI and systems security, where I study how AI systems behave under the real conditions of security-critical applications. My research reveals gaps between benchmark performance and real-world reliability through measurement-driven analysis of deployed systems, and uses these insights to design AI that is robust, adaptive, and secure.
I apply this approach to real systems, including malware detectors and customer-service chatbots, to uncover hidden failure modes and improve AI reliability in practice. My prior work identified the overthinking pathology in deep networks (3, 7, 11); developed realistic, security-grounded threat models for machine learning (2, 3, 4, 5, 9, 10, 15); advanced fairness and privacy techniques (8, 13); and demonstrated how distribution shift undermines deployed defenses (14, 16). My research has been featured in VentureBeat and MIT Technology Review. Most recently, I led a successful Amazon Research Award supporting ongoing work on deployment-aware, adaptable AI for security applications.
I have mentored 20+ undergraduate interns and junior researchers in trustworthy AI (through NSF REU, Summer at MC2, and the ACTION Institute), contributing to multiple top-venue publications (7, 9, 13, 16) and six PhD placements in leading programs.
My name is pronounced Yee-it-JAHN . I usually go by Can, pronounced JAHN, with a soft “J” sound .