Toward a Robust R2D2 Paradigm for Radio-interferometric Imaging: Revisiting Deep Neural Network Training and Architecture

Abstract

The R2D2 Deep Neural Network (DNN) series was recently introduced for image formation in radio interferometry. It can be understood as a learned version of CLEAN, whose minor cycles are substituted with DNNs. We revisit R2D2 on the grounds of series convergence, training methodology, and DNN architecture, improving its robustness in terms of generalizability beyond training conditions, capability to deliver high data fidelity, and epistemic uncertainty. First, while still focusing on telescope-specific training, we enhance the learning process by randomizing Fourier sampling integration times, incorporating multiscan multinoise configurations, and varying imaging settings, including pixel resolution and visibility-weighting scheme. Second, we introduce a convergence criterion whereby the reconstruction process stops when the data residual is compatible with noise, rather than simply using all available DNNs. This not only increases the reconstruction efficiency by reducing its computational cost, but also refines training by pruning out the data/image pairs for which optimal data fidelity is reached before training the next DNN. Third, we substitute R2D2's early U-Net DNN with a novel architecture (U-WDSR) combining U-Net and WDSR, which leverages wide activation, dense skip connections, weight normalization, and low-rank convolution to improve feature reuse and reconstruction precision. As previously, R2D2 was trained for monochromatic intensity imaging with the Very Large Array at fixed $512\times 512$ image size. Simulations on a wide range of inverse problems and a case study on real data reveal that the new R2D2 model consistently outperforms its earlier version in image reconstruction quality, data fidelity, and epistemic uncertainty.

18 pages, 6 figures

References in corpus (21)

• LOFAR: The LOw-Frequency ARray
• WSClean: an implementation of a fast, generic wide-field imager for radio astronomy
• Multi-Scale CLEAN deconvolution of radio synthesis images
• The LOFAR Two-metre Sky Survey -- V. Second data release
• An optimized algorithm for multi-scale wideband deconvolution of radio astronomical images
• Australian Square Kilometre Array Pathfinder: I. System Description
• Sparsity Averaging Reweighted Analysis (SARA): a novel algorithm for radio-interferometric imaging
• Revisiting the radio interferometer measurement equation. II. Calibration and direction-dependent effects
• The MeqTrees software system and its use for third-generation calibration of radio interferometers
• The Planck clusters in the LOFAR sky. I. LoTSS-DR2: new detections and sample overview
• Scalable splitting algorithms for big-data interferometric imaging in the SKA era
• Image reconstruction algorithms in radio interferometry: from handcrafted to learned regularization denoisers
• Comparison of classical and Bayesian imaging in radio interferometry
• An accelerated splitting algorithm for radio-interferometric imaging: when natural and uniform weighting meet
• First AI for deep super-resolution wide-field imaging in radio astronomy: unveiling structure in ESO 137--006
• Bayesian radio interferometric imaging with direction-dependent calibration
• Scalable precision wide-field imaging in radio interferometry: I. uSARA validated on ASKAP data
• ViCTORIA project: The LOFAR HBA Virgo Cluster Survey
• Cygnus A jointly calibrated and imaged via non-convex optimisation from VLA data
• Deep learning-based radiointerferometric imaging with GAN-aided training
• Deep network series for large-scale high-dynamic range imaging