Reproduce “A plug-and-play approach with fast uncertainty quantification for weak lensing mass mapping” on OpenResearch

About this paper

Hubert Leterme, Andreas Tersenov, Jalal Fadili, Jean-Luc Starck

The research introduces PnPMass, a weak lensing mass mapping method that integrates a plug-and-play deep denoiser with a novel, fast uncertainty quantification pipeline. It infers dark matter distribution from noisy shear data, achieving high accuracy and computational efficiency while providing statistically calibrated and tighter per-pixel uncertainty estimates, which are essential for cosmological parameter inference.

Key insights

• A carefully designed plug-and-play framework can decouple the deep denoiser training from observation-specific noise and mask properties, enabling a universal model for diverse weak lensing data.
• Combining moment networks for initial variance prediction with CQR provides statistically robust, non-asymptotic marginal coverage guarantees for uncertainty estimates in mass maps without expensive sampling.
• Prior knowledge about the Gaussian component of mass maps can be effectively integrated into PnP frameworks to slightly enhance reconstruction accuracy, particularly for non-Gaussian features.

Summary by alphaXiv.