Scalable precision wide-field imaging in radio interferometry: II. AIRI validated on ASKAP data

ABSTRACT Accompanying Part I, this sequel delineates a validation of the recently proposed AI for Regularization in radio-interferometric Imaging (AIRI) algorithm on observations from Australian Square Kilometre Array Pathfinder (ASKAP). The monochromatic AIRI-ASKAP images showcased work are formed using same parallelized and automated imaging framework described I: ‘uSARA validated ASKAP data’. Using Plug-and-Play approach, AIRI differs uSARA by substituting trained denoising deep neural network (DNN) proximal operator regularization step forward–backward during deconvolution. We build shelf DNN denoisers that target estimated image dynamic ranges our selected data. Furthermore, we quantify variations reconstructions when selecting nearest versus universal with highest range, opening door to more complete not only delivers estimation but also quantifies epistemic model uncertainty. continue comparative analysis source structure, diffuse flux measurements, spectral index maps sources as imaged algorithms I – WSClean. Overall, see an improvement over WSClean reconstruction components images. scientific potential delivered is evident further precision, accurate maps, significant acceleration deconvolution time, whereby four times faster than its subiterative sparsity-based counterpart uSARA.