Kilo-scale point-source inference using Parametric Cataloging

TL;DR

This paper introduces Parametric Cataloging, a scalable fixed-dimensional probabilistic method for estimating point-source counts in the sky, improving efficiency over traditional trans-dimensional approaches.

Contribution

It presents a novel fixed-dimensional framework for probabilistic cataloging that simplifies implementation and enhances scalability for large astronomical datasets.

Findings

Enables inference of tens of thousands of sources.

Requires only simple reparameterization of flux coordinates.

Compatible with standard gradient-based samplers.

Abstract

The estimation of the number of point-sources in the sky is one the oldest problems in astronomy, yet an easy and efficient method for estimating the uncertainty on these counts is still an open problem. Probabilistic cataloging solves the general point-source inference problem, but the trans-dimensional nature of the inference method requires a bespoke approach that is difficult to scale. Here it is shown that probabilistic cataloging can be performed in a fixed-dimensional framework called Parametric Cataloging under mild assumptions on some of the priors. The method requires only a simple reparameterization of the flux coordinates, yielding an accessible method that can be implemented in most probabilistic programming environments. As the parameter space is fixed-dimensional, off the shelf gradient based samplers can be employed which allows the method to scale to tens of thousands of sources.