Machine Learning Journal Club: ‘Hierarchical Simulation-Based Inference of Supernova Physics’

Speaker: Alex Gagliano (IAIFI/MIT)

Venue&Time: 11:00 / Grey Room 2

Absrtact: Supernovae are powered by diverse physical mechanisms such as radioactive decay, circumstellar interaction, and magnetar spin-down, but distinguishing among them from light curves alone remains a major challenge. In this talk, I’ll present a hierarchical simulation-based inference framework that jointly infers both the dominant power source and the key physical parameters of supernovae. By training conditional neural density estimators end-to-end across multiple physical models, this approach yields calibrated posteriors even in regions of strong degeneracy. I’ll highlight the model’s performance on synthetic data and outline plans to fine-tune the model for population-level studies of diverse explosion physics.