Research
Research directions
- ML X Astrophysics Simulations, forward models, and high-performance methods that connect theory to observations
- Machine Learning for Scientific Discovery Inference in high dimensional spaces, samplers, and principled metrics for accuracy
- Strong Lensing and Dark Matter Mass in galaxies and clusters, substructure, and constraints from multiply imaged systems
ML X Astrophysics
Extracting every bit of information
Our team develops methodology for analyzing astrophysical data with machine learning. We develop machine learning frameworks and apply them to diverse problems including strong gravitational lensing, gravitational waves, stellar populations analysis, extra-solar planet discovieries, causal connections of black holes and their host galaxies, and more.
Machine Learning for Scientific Discovery
Inference in high dimensions
In many physical sciences, posterior sampling is routine in low-dimensional settings. But doing this for large dimensional states (e.g., images) remains an open problems.
Samplers and generative models. A primary focus of our research is building frameworks that fold machine learning model into high-dimensional inference pipelines. These works are related to diffusion models, GFlowNets, normalizing flows, generative modeling, transport maps, or amortized surrogates.
Evaluation and calibration. The second pillar develops metrics and procedures to audit those pipelines: how biased are samples, how faithful are credible regions, and when should we trust a learned approximation? Representative work includes TARP, PQMASS, and MIRA.
Click and drag to rotate the view.
Strong Lensing and Dark Matter
Invisible halos of colossal mass
Our team is a world leader in mapping and interpreting how fragmented and clumpy dark-matter halos of galaxies are. By measuring how smoothly or irregularly that invisible mass is arranged in halos, we aim to constrain particle properties of dark matter that have never been measured directly, sharpen the physical picture of what it could be, and connect those inferences to laboratory searches and broader particle-physics theories. Click and drag to rotate the view.