What do we do?

We use math, physics and computing to find patterns that are hard to see. Current projects:

Why do big models work? Classical statistics says a model with more parameters than data should fail, yet modern machine-learning models often get better as they grow ("double descent"). With Mark Transtrum, Tyler Jarvis and Jared Whitehead we developed the generalized aliasing decomposition, which explains this with linear algebra and tells you how to design a model before you fit it (Phys. Rev. Research, 2025). Good fit if you like linear algebra and want to understand machine learning from the inside.

AI that finds interesting things in images of cells. Cryo-electron tomography makes 3-D images of whole cells. We started with one target, the bacterial flagellar motor: we built MotorBench, about 2,000 tomograms with every motor marked, and ran a Kaggle competition on it in which more than 1,100 people from 70 countries took part and the top teams found about 90% of the motors automatically (Communications Biology, 2026). Now we are building AI that finds interesting structures in tomograms without being told what to look for, by learning the building blocks shared across thousands of them.

Can AI replace quantum mechanics for materials? Universal machine-learned potentials promise DFT accuracy at a tiny fraction of the cost. We tested fourteen of them on 1.45 million alloy structures and showed how much of a "leaderboard" is noise in the reference data itself (arXiv, 2026).

Open-source tools for crystals and alloys. Students help build and maintain Julia packages used around the world: Enumlib.jl (derivative-structure enumeration), Spacey.jl (crystal symmetry) and MinkowskiReduction.jl.

Students co-author papers, present at conferences, and some travel internationally.

What do you have to know?

Students must be motivated to learn and must dependably balance research time with classes, family responsibilities, etc. Funding is available for all students willing to commit to working at least 10 hours/week on average.

Successful students don't usually have prior knowledge in biophysics, strong computational skills, or programming experience. Such experience is nice, but exerting the effort to learn makes all the difference.

Getting Started

Basic skills you'll pick up on the way:

• Linear algebra, statistics, and whichever physics or biology your project needs
• Computational physics and numerical linear algebra
• Programming (some of: julia, python, C++, Fortran)
• Using the command line and scripting
• Using git and GitHub
• Using BYU's supercomputer
• Using a text editor (emacs, vi, etc.) and/or an "IDE"—integrated development environment (vscode, sublime, etc.)

If you want to see if you would enjoy working in our group,

• Work through the "getting started" tutorial
• Work through a unix tutorial (on a mac, this is easy, since macOS is unix-based). If you have a Windows machine, use WSL (Windows Subsystem for Linux).
• Try learning a little julia
• Come to "Tea and Chat" and talk to students in the group (Mondays at 4 pm in N288 Eyring Science Center)