You can find details about current and recent projects below.

Bayesian Shear Inference with Differentiable Forward Modeling

  • We leverage the differentiable forward model of galaxies from JAX-GalSim, gradient based samplers, GPUs, and Schneider et al.’s importance sampling approach to develop a new efficient Bayesian algorithm for measuring cosmic shear.

pgm

Differentiable Forward Models of Galaxy Light Profiles

  • I lead the development JAX-GalSim, a GPU-accelerated and differentiable version of GalSim, which is currently under active development.

Machine Learning models for mitigating the galaxy-galaxy blending problem in cosmological surveys

  • We developed BLISS a machine learning model for probablistic inference of galaxy properties specifically targeted at blended galaxy fields.

bliss

This figure showcases the overall structure of the BLISS inference pipeline.

Statistical framework for Galaxy-Halo Connection on N-body Simulations

  • I developed MultiCAM, a multi-variable extension to conditional abundance matching (CAM) that can be used to connect properties of dark matter haloes with properties of galaxies.

multicam

Above is a comparison of the correlation strength between predictions of MultiCAM and CAM, where MultiCAM can use the full mass accretion history (MAH) of a dark matter haloes as features for prediction.

Framework for evaluating galaxy deblending algorithms

  • I lead the development of BTK a software tool for simulating galaxy blends and consistent comparing galaxy deblenders based onv various metrics.

btk

This figure compares the reconstruction quality of individual galaxies from SExtractor and Scarlet when applied to simulated galaxy blends produced by BTK. The three metrics have been previously used in galaxy blending literature.