Modern Approaches to Radio Data Analysis

Radio astronomy is experiencing a renaissance, with new low-frequency datasets potentially exceeding modern cosmological datasets by a factor of ~1000 in size. It is thus timely to push on the data frontier of radio astronomy, and in this talk I will describe two advances in this area.
First, we ask the most fundamental question of survey science: what does the sky look like in all directions at all frequencies? Answering this question is non-trivial, as our knowledge of the sky is formed from a patchwork of surveys that are incomplete in both frequency and sky coverage. I will describe an ongoing effort to use advanced statistical techniques to combine all existing archival data into a single statistical model that allows maps of diffuse radio emission to be predicted at any frequency between 10 MHz and 5 THz.
Moving beyond sky surveys to constraints in high-redshift radio astronomy, I will then show that cosmology and radio astronomy are now inextricably linked: radio surveys will soon be sufficiently precise that cosmological parameter variations must be included when fitting low-frequency radio data to models of cosmic dawn (i.e., the epoch of first stars and galaxies). Unfortunately, this presents a computational challenge for parameter space explorations, since detailed simulations of cosmic dawn are extremely slow to run. I will describe how machine learning algorithms can be employed to speed up the model-fitting process. Our code pushes the state-of-the-art in cosmic dawn parameter fits from a 4-dimensional parameter space of only astrophysical parameters to an 11-dimensional parameter space of astrophysical and cosmological parameters. This represents a crucial step in the establishment of low-frequency radio astronomy as a workhorse probe of both astrophysics and cosmology.