Hudson Smith

I am an Assistant Professor in the School of Mathematical and Statistical Sciences at Clemson University and a co-lead of the Clemson/MUSC Artificial Intelligence Hub. I develop machine learning methods that incorporate prior knowledge for data-constrained scientific and medical applications.

The synthesis of established domain knowledge with flexible learning-based methods leads to better anomaly detection, improved sample efficiency, and more explainable inference without sacrificing the expressive power of modern data-driven models.

My group’s current work runs along three threads: making model predictions interpretable and trustworthy enough for clinical use, scaling exact Bayesian and point-process inference to datasets that were previously out of reach, and building medical imaging models that learn from weak or noisy labels. See my publications.

I earned a PhD in theoretical atomic physics from The Ohio State University. As a physicist, I learned the value of first-principles reasoning; as a machine learning researcher, I have seen the power of flexible statistical models. That combination shapes how I approach research. See my physics-based visualizations.