CAIS Talks with Azadeh Tabiban
The Centre for Artificial Intelligence & Society presents:
CAIS Talks with Azadeh Tabiban
Date: Thursday, November 12, 2026
Time: 11:00am-12:30pm EST
Location: Kaneff Tower 519
On the Robustness of Learning-based Intrusion Detectors
Abstract:
Provenance-based intrusion detection systems (PIDSes) are a promising approach for detecting sophisticated attacks. Learning-based PIDSes learn from the captured causal relationships between events to identify malicious system entities based on their context. However, can these detectors remain effective when attackers surround malicious entities with carefully chosen benign-looking events? In this talk, I will present Contorter, an evasion framework that leverages node-level PIDS embeddings to guide the generation of effective benign-looking events. Contorter identifies benign entities with contexts similar to those of malicious entities and replicates their interactions around the malicious ones to evade detection. I will then discuss how comparing local and broader contexts and examining rare events can help uncover concealed malicious behavior. Finally, I will outline open questions about detector robustness and directions for future research.
Bio:
Azadeh Tabiban is an Assistant Professor in the Department of Computer Science at the University of Manitoba, where she leads the FOCUS research lab. Her research focuses on evasion attacks and robust security, network security, machine learning applied to security, provenance analysis and forensics, and cloud/edge security. Her work has appeared in leading security venues, including IEEE Symposium on Security and Privacy and NDSS. Previously, she was an industrial postdoctoral fellow at the University of Waterloo, collaborating with Ericsson Montreal. She received her PhD from Concordia University.

