Rick S. Blum, the Robert W. Wieseman Endowed Professor in Electrical Engineering, is presenting several invited lectures on cybersecurity, AI, and sensing systems this fall.

He has been invited to give a keynote at the 2025 IEEE International Carnahan Conference on Security Technology in San Antonio, Texas, from Oct. 13–17. The talk, titled “Cyber Security of Sensor Systems for State Sequence Estimation: An AI Approach,” will focus on methods to detect and protect against attacks on sensor data using machine learning approaches. On Oct. 14, Blum will present the same talk at Trinity University in San Antonio.

He also will give an invited lecture at Arizona State University on Nov. 7, titled “Explainable Anomaly Detection in Dynamical Systems to Extract More Information and Statistical Shapley Value Analysis,” which introduces a new machine learning model for identifying, classifying, and understanding anomalies in dynamical systems.

Earlier in October, Blum will travel to Finland (Oct. 1–4) to serve as the Opponent for a PhD defense at Aalto University in Helsinki. In the European tradition, the Opponent is an expert who critiques the dissertation, asks the candidate questions, and evaluates whether the research meets the standards for a PhD. Blum’s research—highly cited work in radar, communications, and MIMO radar—makes him well qualified to evaluate the dissertation on Integrated Sensing and Communications (ISAC), a new approach for 6G cellular networks. ISAC has the potential to enhance applications such as autonomous vehicles and remote patient monitoring.

If the student passes, a celebration follows immediately afterward, attended by the candidate’s family and friends. The event is both rigorous and festive, combining formal examination with a large gathering that Blum likens to a wedding party. 

Read abstracts of the talks below.


Cyber Security of Sensor Systems for State Sequence Estimation: An AI Approach
Due to possible devastating consequences, counteracting sensor data attacks is an extremely important topic, which has not seen sufficient study. This presentation presents the first methods that accurately identify/eliminate only the problematic attacked sensor data presented to a sequence estimation/regression algorithm under a powerful attack model. The approach does not assume a known form for the statistical model of the sensor data, allowing data-driven and machine learning sequence estimation/regression algorithms to be protected. A simple protection approach for attackers not endowed with knowledge of the details of our protection approach is first developed, followed by additional processing for attacks based on protection system knowledge. Experimental results show that the simple approach achieves performance indistinguishable from that for an approach which knows which sensors are attacked. For cases where the attacker has knowledge of the protection approach, experimental results indicate the additional processing can be configured so that the worst-case degradation under the additional processing and a large number of sensors attacked can be made significantly smaller than the worst-case degradation of the simple approach, and close to an approach which knows which sensors are attacked, with just a slight degradation under no attacks. Mathematical descriptions of the worst-case attacks are used to demonstrate the additional processing will provide similar advantages for cases for which we do not have numerical results. All the data-driven processing used in our approaches employs only unattacked training data.
 

Explainable Anomaly Detection in Dynamical Systems to Extract More Information and Statistical Shapley Value Analysis

It is common to monitor physical systems, whose dynamics typically follow ordinary differential equations (ODEs), using a set of sensor observations. Anomaly detection is an important technique which attempts to determine if there is something is wrong with the sensor data. Improving the performance of anomaly detection is important and we will provide improved anomaly detection approaches, but we want to go much further than this and attempt to extract much more information about the anomaly which can help determine the exact problem, what is the cause of the problem, how soon one needs to act on this problem, and exactly what actions are most appropriate to resolve any important issues related to the anomaly. In our preliminary results, we demonstrate we have a new approach which improves the performance of anomaly detection and root cause analysis while, for the first time, providing critical additional information in the form of anomaly classification, which describes the exact type of anomaly that occurred. This is accomplished using a new machine learning model that we developed called Interpretable Causality Ordinary Differential Equation (ICODE) which exploits Neural ODE ideas. We provide a theoretical analysis of how to detect, localize, find the root cause of an anomaly and classify anomaly type in ODE data using ICODE. This analysis explains ICODE’s excellent performance via its ability to recognize after-anomaly changes in the causal graph describing the variables in the differential equation. We provide experiments on three simulated ODE systems that demonstrate ICODE’s superior performance in anomaly detection, root cause localization, and type classification. Here we consider two types of anomalies, cyber anomalies and measurement anomalies which we described mathematically for the first time. Cyber anomalies occur when the values of the variables in the ODE representing the real physical system are changed, typically due to the physical system being damaged. Measurement anomalies occur when only the sensor data are modified while the real physical system is unchanged. In part 2 of this talk we illustrate some bad behavior when we statistically analyze using the Shapley value for anomaly localization.