I build probabilistic, data-driven models of intrinsic stellar colors and combine them with Gaia distances and multi-band photometric surveys to infer extinction and stellar parameters in diffuse, high-latitude regions. The focus is calibrated uncertainty and robust behavior where classical assumptions can fail.
I develop SBI workflows that remain stable under model misspecification by learning informative summary statistics, checking posterior calibration, and explicitly tracking simulator-observation mismatch. The goal is reliable uncertainty quantification for scientific decision-making.
I am interested in combining BayeSN with lensing-aware inference to constrain cosmological parameters, including H0, from Type Ia supernovae. This line of work connects principled generative modeling with precision cosmology.
Myself, Gerry Gilmore and Kaisey Mandel are developing a method to learn the colour-magnitude diagram of low-extinction stars at high Galactic latitudes using normalzing flows.
I have developed normalizing architectures and shown that under certain assumptions the No U-turn Sampler will converge to the target density if it constains a normalizing flow component in the Hamiltonian. In the GIF below I show the leap-frog integration for a Hamiltonian with normalizing flow component. 