SIFARI: Self-Supervised Interferometric Fitting for Astronomical Radio Imaging
Authors: Shunyuan Mao, Andrea Isella, Paris Perdikaris, Li-Ta Lo, Hui Li
Organizations: Department of Physics and Astronomy, Rice University, 6100 Main St, Houston, TX 77005, USA · Rice Space Institute, Rice University, 6100 Main St, Houston, TX 77005, USA · National Radio Astronomy Observatory, 520 Edgemont Road, Charlottesville, VA 22903, USA · NSF-Simons AI Institute for Cosmic Origins, Austin, TX 78712, USA · Department of Mechanical Engineering and Applied Mechanics, University of Pennsylvania, 220 S 33rd St, Philadelphia, PA 19104, USA · Computing and Artificial Intelligence Division, Los Alamos National Laboratory, Los Alamos, NM 87545, USA · Theoretical Division, Los Alamos National Laboratory, Los Alamos, NM 87545, USA
Radio-interferometric images are reconstructed from sparsely sampled visibilities, and CLEAN-based imaging can struggle with spatial filtering, complex morphologies, and uncertainty quantification. Alternative methods that fit visibilities directly can address some of these limitations but often require manual choices of image priors and model hyperparameters. We present SIFARI (Self-Supervised Interferometric Fitting for Astronomical Radio Imaging), a self-supervised neural network workflow that represents sky brightness as a continuous function of position and fits measured visibilities without an external image training set or explicit spatial regularizer. An empirical rule sets the Fourier feature scale from the visibilities before training, controlling how readily the network fits fine structure. Sampling network weights with Stochastic Weight Averaging-Gaussian (SWAG) gives approximate brightness uncertainty estimates, which we combine with a thermal-noise floor to construct spatially resolved signal-to-noise maps. In synthetic ALMA tests, SIFARI yields an effective point-source response about eight times narrower than the natural-weighting CLEAN restoring beam and recovers more extended flux than CLEAN when short baselines are missing. It also achieves higher image fidelity than the restored CLEAN images in all three morphology benchmarks. Applied to ALMA observations of PDS 70, SIFARI recovers the bright outer ring together with faint compact emission in the central cavity. For long-baseline-only WISPIT 2 data, SIFARI supplies a sky model for phase self-calibration where the CLEAN model is inadequate. The restored, self-calibrated SIFARI image has approximately 30% lower RMS noise than the CLEAN image made from the original visibilities without self-calibration.
Figures & tables
Figure 1: Two-stage SIFARI workflow for visibility-domain image reconstruction and uncertainty quantification. A coordinate-based neural network maps sky coordinates (x,y) to brightness I(x,y) . Before training, the visibility data set the frequency scale σν used to choose νk (dashed arrow). The model image is Fourier-transformed at the observed (ul,vl) coordinates to predict visibilities, which are compared with the data through the likelihood. The weights Wl are inverse noise variances of the real and imaginary visibility components. Stage 1 uses Muon+AdamW to fit the network and yields the reported best-fitting reconstruction. Stage 2 starts from this fit, uses SGD to build the SWAG approximation, and samples network weights to estimate the pixel-wise brightness uncertainty.
Category
Hyperparameter
Value
Architecture
Neurons per hidden layer M
128
Number of hidden layers
2
Hidden-layer activation
ELU
Fourier feature scale σν
Estimated from visibilities
Number of Fourier features K
512
Stage 1: Optimization
Initial Muon learning rate
2×10−2
Table 1: Default hyperparameters used in all experiments unless otherwise specified. Synthetic models and their observing setups are listed in Table 2 .
Source Model
ALMA configurations used
Ground-truth pixels per side
Ground-truth pixel size (″)
One point source
C-3 (20 min) + C-5 (30 min) + C-9 (60 min)
1024
0.0005
Two point sources ( ΔRA=0.009′′ )
C-3 (20 min) + C-5 (30 min) + C-9 (60 min)
1024
0.0005
Two point sources ( ΔRA=0.010′′ )
Uniform disk ( θUD=0.2\arcsec )
C-9 (60 min)
200
0.0015
Uniform disk ( θUD=0.6\arcsec )
680
0.0015
Uniform disk ( θUD=1.0\arcsec )
960
0.0015
Table 2: Synthetic benchmark models and their ALMA simobserve setups. The grid columns specify the ground-truth images used to generate the synthetic visibilities; reconstruction grids are specified separately in the text. The C-3, C-5, and C-9 configurations span baseline ranges of 15–500 m, 15–1398 m, and 368–13 894 m, respectively. Rows with multiple configurations use their combined visibilities.
Figure 2: Normalized 1D slices along RA of the dirty beams (gray) and the SIFARI point-source response (red). Natural, Briggs, and uniform weighting are shown with dashed, dash-dotted, and dotted curves, respectively; SIFARI is shown with a solid curve. The Briggs dirty beam uses a robust parameter of 0.5. Each profile is normalized to its peak. The vertical dashed lines mark ±λ/Bmax .
Figure 3: Two-point-source tests: normalized 1D flux slices of CLEAN and SIFARI reconstructions along the RA axis connecting the sources (linear scale). Natural-, Briggs-, and uniform-weighted CLEAN images are shown with gray dashed, dash-dotted, and dotted curves, respectively; SIFARI is shown with a red solid curve. Each profile is normalized to its peak. The vertical dashed lines mark the injected source positions. Left: ΔRA=0.009′′ . Right: ΔRA=0.010′′ .
Disk diameter
Model Flux
CLEAN
SIFARI
(″)
(mJy)
(mJy)
(mJy)
0.2
200
200
200
0.6
200
41
202
1.0
200
20
219
Table 3: Flux recovery for uniform-disk models observed with the ALMA C-9 configuration. Recovered fluxes are measured within a circular aperture with the same diameter as the corresponding disk. The CLEAN images use natural weighting.
Figure 5: CLEAN and SIFARI reconstructions of the three models described in Sec. III.4 . Rows show the inclined ring with a point source, the multiple-ring system, and the molecular cloud, from top to bottom. Columns show the ground truth, CLEAN model, restored CLEAN image, and SIFARI reconstruction, from left to right. Each row uses a common logarithmic intensity scale. In the inclined-ring model (top row), the cyan plus sign marks the injected point-source position. The three numbers above each reconstruction give IFS, SSIM, and the recovered total-flux ratio, respectively. The FWHM of the CLEAN beam is 0.032′′×0.027′′ .
Figure 6: SIFARI reconstructions and uncertainty diagnostics for the three morphology benchmarks in Figure 5 . Rows show the inclined ring with a point source, the multiple-ring system, and the molecular cloud, from top to bottom. Columns show the SIFARI reconstruction, the pixel-wise standard deviation across SWAG samples ( σSWAG ), and the SNR defined by Eq. 7 , from left to right. The white contours on the reconstruction and SNR panels mark SNR=5.
Figure 7: Comparison of reconstructed images and signal-to-noise ratio (SNR) diagnostics for real PDS 70 data. Top: global view. Bottom: zoomed-in view of the central region, including the central emission and the locations of PDS 70 b and PDS 70 c. In each row, the panels show, from left to right, the restored CLEAN image, the SIFARI reconstruction, and the SIFARI SNR map. The CLEAN and SIFARI images share a common intensity scale. The white contours on the SIFARI and SNR panels mark SNR=5, as defined in Eq. 7 . The CLEAN beam is 0.045′′×0.034′′ .
Figure 8: Three image reconstructions of the WISPIT 2 continuum, shown with a common color scale. Left: naturally weighted CLEAN image from the original calibrated visibilities, before self-calibration. Middle: SIFARI model fitted to the self-calibrated visibilities, without beam convolution or added residuals. Right: restored SIFARI image, formed by convolving the middle-panel model with the restoring beam and adding the residual image from the self-calibrated visibilities.
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
Figure 9: Comparison of SIFARI reconstructions for different Fourier feature scales. Rows (top to bottom) show the inclined-ring, multiple-ring, and molecular-cloud benchmarks, with a common logarithmic intensity scale within each row. The leftmost column shows the ground truth; the remaining panels show reconstructions at the Fourier feature scales σν indicated in their titles. The scale selected by the visibility-supported heuristic estimator is marked “(auto).” Panel-header annotations report image fidelity (IFS; left), SSIM (center), and recovered flux ratio (right).
The detection limit of astronomical imaging observations is limited by several noise sources. Some of that noise is correlated between neighbouring image pixels and exposures, so in principle could be learned and corrected. We present an astronomical self-supervised transformer-based denoising algorithm (ASTERIS), that integrates spatiotemporal information across multiple exposures. Benchmarking on mock data indicates that ASTERIS improves detection limits by 1.0 magnitude at 90% completeness and purity, while preserving the point spread function and photometric accuracy. Observational validation using data from the James Webb Space Telescope (JWST) and Subaru telescope identifies previously undetectable features, including low-surface-brightness galaxy structures and gravitationally-lensed arcs. Applied to deep JWST images, ASTERIS identifies three times more redshift > 9 galaxy candidates, with rest-frame ultraviolet luminosity 1.0 magnitude fainter, than previous methods.
Yuduo Guo, Hao Zhang, Mingyu Li +10
Department of Automation, Beijing National Research Center for Information Science and Technology, Tsinghua University, Beijing, China · Institute for Brain and Cognitive Sciences, Tsinghua University, Beijing, China · Department of Astronomy, Tsinghua University, Beijing, China +6
Radio interferometry enables high-resolution imaging of astronomical radio sources by synthesizing a large effective aperture from an array of antennas and solving a deconvolution problem to reconstruct the image. Deep learning has emerged as a promising solution to the imaging problem, reducing computational costs and enabling super-resolution. However, existing DL-based methods often fall short of the requirements for real-world deployment due to limitations in handling high dynamic range, large field of view, and mismatches between training and test conditions. In this work, we build upon and extend the POLISH framework, a recent DL model for radio interferometric imaging. We introduce key improvements to enable robust reconstruction and super-resolution under real-world conditions: (1) a patch-wise training and stitching strategy for scaling to wide-field imaging and (2) a nonlinear arcsinh-based intensity transformation to manage high dynamic range. We conduct comprehensive evaluations using the T-RECS simulation suite with realistic sky models and point spead functions (PSF), and demonstrate that our approach significantly improves reconstruction quality and robustness. We test the model on realistic simulated strong gravitational lenses and show that lens systems with Einstein radii near the PSF scale can be recovered after deconvolution with our POLISH model, potentially yielding 10× more galaxy-galaxy lensing systems from the Deep Synoptic Array (DSA) survey than with image-plane CLEAN. Our results highlight the potential of DL models as practical, scalable tools for next-generation radio astronomy.
Zihui Wu, Liam Connor, Samuel McCarty +1
Department of Computing and Mathematical Sciences, California Institute of Technology (Caltech), Pasadena, CA, USA · Center for Astrophysics | Harvard & Smithsonian, Cambridge, MA, USA
Radio astronomy plays a crucial role in understanding the universe, particularly within the realm of non-thermal astrophysics. Images of celestial objects are derived from the signals (called visibility) measured by radio telescopes. Such imaging results, called dirty images, contain artifacts due to factors such as sparsity and therefore require reconstruction to improve imaging quality. Existing methods typically restrict reconstruction to a unimodal domain, either to the dirty image after imaging or to the sparse visibility prior to imaging. Focusing solely on each unimodal reconstruction results in the loss of complementary in-context information in either the visibility or image domain, leading to an incomplete modeling of mutual dependency and consistency. To address these challenges, we propose CDCRec, a multimodal radio interferometric data reconstruction method that explicitly models cross-domain consistency. We design a hierarchical multi-task and multi-stage framework to enhance the exploration of interplays between domains during reconstruction. Our experimental results demonstrate that CDCRec improves imaging performance through enhanced cross-domain correlation extraction. In particular, our self-supervised complementary modeling strategy is better than current methods at interferometric domain translations that rely heavily on recovering dense information from constrained source-domain data.
Kai Cheng, Ruoqi Wang, Qiong Luo
The Hong Kong University of Science and Technology · The Hong Kong University of Science and Technology (Guangzhou)