a957@phase-space:~$ biology (dynamical system)
Research
Current work sits where cellular state, dynamical systems, and scientific machine learning overlap.
theme 01
I am drawn to the question of how cells commit. The language I use is dynamical systems — attractors, basins of attraction, and vector fields — applied to RNA expression profiles. The goal is a description of cellular decision-making that is richer than a cluster label: one that says something about the forces at work, the stabilities available, and the transitions that remain possible.
theme 02
At JAX I contribute to tools that make automated mouse behavior analysis practical and genetically informative. This work involves the full stack — video, pose, classifier, genetics — and the recurring challenge is designing systems that scientists actually want to use.
theme 03
My earlier work asked whether knowing the structure of a physical system — its conservation laws, its symmetries, its phase-space geometry — makes a neural network both more accurate and more honest. The answer was yes, often dramatically so. I continue to think about what it means to build models that remain faithful to mechanism.
Selected publications
A practical, open platform for data acquisition, behavior annotation, classifier sharing, and downstream genetic analysis in mouse studies.
Shows that learned neuronal diversity can improve model performance in nonlinear regression and physics-informed prediction tasks.
Uses Hamiltonian structure to learn phase-space orbits even as nonlinear systems transition between regular and chaotic regimes.
Develops a stability framework for understanding how simultaneous perturbations affect resilience in multistable networked systems.