26th Australian Institute of Physics Congress and the Australian and New Zealand Conference on Optics and Photonics

Differentiable Models for Radio Astronomy Interferometer Calibration and Image Synthesis

Not scheduled

20m

Belinda Hutchinson Building (The University of Sydney )

Contributed Oral AIP | Astronomy (ASTRO)

Description

Astronomical radio interferometers achieve exquisite angular resolution by cross-correlating signals from a cosmic source simultaneously observed by distant pairs of antennas to produce a Fourier-type measurement called a visibility. Here we describe the implementation of a feed-forward modeling approach to synthesize scientific images from interferometric visibility datasets, built using the radio interferometric measurement equation and the autodifferentiable machine learning framework PyTorch, called MPoL. Neural network components provide a rich set of modular and composable building blocks that can be used to express the physical relationships between latent model parameters and observed data, including parameters which might otherwise be restricted to a one-time calibration procedure prior to image synthesis. Industry-grade optimizers make it straightforward to simultaneously solve for the synthesized image and calibration parameters using stochastic gradient descent. As a demonstration, we apply this framework to the ALMA DSHARP dataset of the protoplanetary disk hosted by IM Lup and achieve a 30\% improvement in spatial resolution and a 65\% improvement in peak brightness compared to the standard CLEAN imaging procedure.

https://mpol-dev.github.io/MPoL/