Machine learning & AI for data with complex structure.
My research develops machine learning and artificial intelligence methods for data whose relationships matter as much as the data points themselves — centered on graph and hypergraph neural networks and geometric deep learning, with applied work in AI for 3D vision and computational imaging.
Learning on graphs, hypergraphs, and beyond
Much of the world's most important data — power grids, brain connectomes, molecules, sensor networks — is fundamentally relational. My work builds models that learn directly on those structures rather than flattening them away.
Graph & hypergraph neural networks
Tensor-based hypergraph neural networks (T-HyperGNNs, HyperNATE) and hypergraph U-Net architectures that capture higher-order relationships beyond pairwise edges.
Geometric deep learning
Deep models that respect the geometry and symmetry of non-Euclidean domains, generalizing convolution and attention to graphs and hypergraphs.
Graph signal processing
Sampling, filtering, and spectral analysis of signals defined on graphs and multigraphs — including blue-noise sampling on non-Euclidean domains.
Graph learning for physical systems
Inferring structure and state in power-distribution networks and neural connectomes — phase identification, consensus graph learning, and time-series clustering.
AI for 3D vision
Deep learning applied to structured-light 3D capture and computational imaging — processing 3D video for real-world machine-vision systems.
Computational imaging
Recovering structure and spectral detail beyond a single exposure, including multispectral and hyperspectral imaging methods.
Selected recent work
A representative slice of the graph- and hypergraph-learning research. See Google Scholar for the full record.
T-HyperGNNs: Hypergraph Neural Networks via Tensor Representations
A tensor framework for hypergraph neural networks that models higher-order relationships without lossy clique expansion. doi.org/10.1109/TNNLS.2024.3371382 →
HyperNATE: Scaling Tensor-Based Hypergraph NNs Through Attention
An attention mechanism that scales tensor-based hypergraph neural networks to large, real-world hypergraphs. doi.org/10.1016/j.neunet.2026.109139 →
Blue-Noise Sampling of Graph & Multigraph Signals
Dithering on non-Euclidean domains — extending blue-noise sampling theory from images to signals defined on graphs and multigraphs.
t-HGSP: Hypergraph Signal Processing via t-Product Tensor Decompositions
A signal-processing framework for hypergraphs built on the t-product tensor decomposition, foundational to the T-HyperGNN line of work.
Scalable Hypergraph Structure Learning with Diverse Smoothness Priors
Learning hypergraph structure from data at scale under a family of smoothness priors that capture different notions of higher-order regularity.
Full publication record
120+ publications spanning graph & hypergraph learning, computational imaging, and 3D vision.
Where the methods land
The same imaging and signal-processing research underpins issued patents and shipping software.
40+ patents
Structured-light 3D scanning, computational and hyperspectral imaging, and digital halftoning. Browse the patents →
FM Halftone Screening for Photoshop
An applied product of the halftoning and dot-placement research — production-grade stochastic screening inside Photoshop. See the plugin →
Sponsored research
This research has been supported by the National Science Foundation (NSF), the Air Force Office of Scientific Research (AFOSR), the Department of Energy (DOE), and industry partners.
Collaboration & affiliation
Databeam Professor of Electrical & Computer Engineering and Director of Graduate Studies at the University of Kentucky. Open to collaborations in graph/hypergraph machine learning, geometric deep learning, and AI for imaging.
Interested in graph & hypergraph machine learning?
Reach out about collaborations, students, or applying these methods to your data.
Contact Dr. Lau →