Authors:

Braden Garretson

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Upcoming all sky surveys, such as the Vera Rubin Observatory, will produce terabytes of data per night and provide the unique opportunity to conduct population scale studies of supernovae and their host galaxy environments. However, this large influx of data also presents numerous technical challenges. Namely, that estimating model posteriors on this scale using traditional sampling techniques, such as Approximate Bayesian Computation (ABC), is infeasible. By replacing traditional ABC methods with amortized posterior inference we are able to train a neural network to estimate accurate posteriors of supernova model parameters at a fraction of the computational cost, and apply this to over 10,000 supernova-like light curves. Using this, we will conduct the largest supernova host galaxy population study to date by empirically measuring whether or not the distribution functions of the supernova model parameters have any dependence on their host galaxy properties. This approach of adapting physical models to neural networks will be essential to rapidly estimate the posterior of model parameters and conduct population scale studies of supernovae and their host galaxy environments in future all-sky surveys.

Source:

Purdue University / 2023

Topics:

No topics listed

Co-authors:

Braden Garretson