About

Hi there, I’m Mattson.

I’m a CS PhD candidate at Northwestern University advised by Han Liu, and I work on machine learning for scientific discovery. To this end, I split my time between exploratory research and real-world ML integration.

My research looks at how we can learn interpretable graph structures directly from data without relying on heuristics that simplify (or, in most cases, obfuscate) the structure-learning signal. Structure learning is, in my opinion, a hugely important and undervalued research area, as graph structure is the foundation upon which all the recent successes of graph neural networks implicitly rely.

Additionally, the utility of most ML models extends well beyond their academic usage, and by working from first principles we can adapt and integrate techniques from apparently disparate domains to aid in the scientific process. In collaboration with Fermilab, I’ve transformed language models into proton-beam extractors and repurposed biomedical image segmentation models to monitor and control particle accelerators. Together, we’ve published four papers on our work and are in the process of implementing these ML systems on FPGA for real-time intervention in the accelerator complex.

Some select projects in manufacturing, organic materials research, particle physics, medical imaging, quant finance and social good are highlighted below. If any of these interest you as well, I’d love to chat.

For more general info, see my brief bio, CV, LinkedIn, GitHub, or Publications.

Select Projects

Graph Structure Learning

Manufacturing and physical control

Optical and Electronic Properties of Organic Materials

Quantitative Finance

Large-scale ML runtime optimization

High Performance Computing in Healthcare

AI for social good

Deep Learning Seminars

Publications

* Equal contribution.

Industrial White Papers