Research themes
Cell fate & transcriptomic dynamics
How does a cell commit to what it becomes? I think about this with the tools of dynamical systems (attractors, basins, vector fields) applied to RNA expression. A cluster label tells you where a cell sits. I want a picture that also says which way it is being pushed, and which fates are still open to it (and how hard it would be to get there).
Behavioral phenotyping & machine learning
At JAX I led JABS (the JAX Animal Behavior System, with Vivek Kumar's lab). It is an open platform that takes you from video of a mouse to its pose, then to behavior classifiers, and finally to genetics. A lot of the work is in the plumbing between those steps, and the part I care about most is making it something other labs actually pick up and use.
Physics-aware scientific machine learning
At NC State (with William Ditto, John Lindner and Sudeshna Sinha) I asked a simple question: if a neural network knows the structure of the physics (energy conservation, the geometry of phase space), does it forecast better? For Hamiltonian systems it mostly did, and the gain was largest where ordinary networks get lost (through the order-to-chaos transition and across separatrices). We also looked at how this scales with data and dimension, and found that letting neurons differ from each other helps learning too.
Dynamics on complex networks
My PhD (at IISER Mohali) and first postdoc (at Oldenburg) were about dynamics on networks, mostly coupled oscillators. The wiring (fixed, rewired over time, or small-world) decides whether they fall into step. I spent a lot of time on how far a synchronized network can be pushed before it stops coming back (basin stability for many nodes at once, and recovery times after local shocks). At Oldenburg I moved to synchronization in ecological food webs, and to a curious case where weakly coupled units lock in phase while the strongly coupled ones keep drifting.