Join GW Engineering's Department of Computer Science for the first installment in their Spring 2026 Colloquium Series! The talk titled "Mind the Gap: Improving the Effectiveness of Machine Learning for Industrial Control Systems Security" will be given by Prof. Clement Fund from Carnegie Mellon University!

Abstract 

Industrial control systems (ICS) govern processes in critical infrastructure, such as power generation, chemical processing, and water treatment. To defend ICS from attacks, a common research proposal is to use machine learning (ML) to detect anomalies in process data, but ML is rarely adopted for ICS in practice today. In this talk, I cover work that makes ML more effective for ICS security, both by investigating needs and opportunities in practice and by developing new ML-based approaches to meet these opportunities. First, to better understand how ML could be used for ICS in practice, we interview practitioners that work in ICS security and operations to understand the requirements for adopting ML into ICS environments. Second, after identifying a clear need for explanations with detected anomalies; we adapt and evaluate a variety of explainable AI (XAI) approaches for attributing ICS anomalies to the underlying components that were manipulated. We ultimately propose an ensemble-based XAI method that can diagnose anomalies 35% faster than individual methods. Third, after identifying a need for more efficient, ICS-specific models, we design CyPRESS, a new model architecture for ICS anomaly detection. CyPRESS focuses on learning the feature relationships that matter for ICS; it is more efficient (by a factor of 1000x), improves attribution, and is more robust to evasion attacks than deep-learning models. Finally, I conclude with my plans for near-term future work and my long-term vision for bridging the gap between ML research and critical infrastructure security.

Bio

Clement Fung is a postdoctoral research associate at Carnegie Mellon University and a member of the CyLab security and privacy institute. His research interests are at the intersection of security, machine learning, and cyber-physical systems. In particular, he focuses on making machine learning more effective at securing industrial control systems in practice. He obtained his Ph.D in Societal Computing from Carnegie Mellon University, his M.S. in Computer Science from the University of British Columbia, and his B.A.Sc. in Systems Design Engineering from the University of Waterloo.