Research
Three threads, in rough order of how much of my time they have taken.
Graph learning for power-system security
State estimation gives a grid operator its picture of the network. A false data injection attack edits the measurements feeding that estimate so the result is wrong but internally consistent, which is exactly what defeats a residual-based bad-data detector. Treating the system as the graph it is — buses and branches, with measurements attached to them — lets a model use topology and the correlations it implies, rather than treating each measurement as an independent number. That line of work covers detection, and then localization: not just that something was tampered with, but where.
Graph neural network methods
Working on the applications above meant working on the models: filter design for graph convolutions, including infinite impulse response and Chebyshev formulations, and the practical question of what makes deeper graph networks trainable. Also the problem of test data — realistic cyber-physical network topologies are scarce, and generating synthetic ones that are connected, simple, and plausible is its own task.
Measurement in craniofacial surgery
A separate collaboration, on quantifying how far a face departs from typical. The clinical problem is that surgical outcomes are usually graded by eye; a learned, personalized measure gives something comparable across patients and over time.
Selected work
Joint detection and localization of stealth false data injection attacks in smart grids using graph neural networks
Graph neural networks based detection of stealth false data injection attacks in smart grids
Generalized graph neural network-based detection of false data injection attacks in smart grids
Spatio-temporal failure propagation in cyber-physical power systems
Personalized quantification of facial normality
Graph Neural Network-Based Cybersecurity of Smart Grids
The full list is on Google Scholar.