Uncertainty quantification for radio interferometric imaging – I. Proximal MCMC methods

• DOI: 10.1093/mnras/sty2004
• Publication Date: 2018-07-26T07:16:28Z
• Authors: Xiaohao Cai; Marcelo Pereyra; Jason D McEwen
• Keywords: Uncertainty Quantification

Uncertainty quantification is a critical missing component in radio interferometric imaging that will only become increasingly important as the big-data era of interferometry emerges. Since requires solving high-dimensional, ill-posed inverse problem, uncertainty difficult but also to accurate scientific interpretation observations. Statistical sampling approaches perform Bayesian inference, like Markov Chain Monte Carlo (MCMC) sampling, can principle recover full posterior distribution image, from which uncertainties then be quantified. However, traditional high-dimensional methods are generally limited smooth (e.g. Gaussian) priors and cannot used with sparsity-promoting priors. Sparse priors, motivated by theory compressive sensing, have been shown highly effective for imaging. In this article proximal MCMC developed imaging, leveraging calculus support non-differential such sparse framework. Furthermore, three strategies quantify using recovered developed: (i) local (pixel-wise) credible intervals provide error bars each individual pixel; (ii) highest density regions; (iii) hypothesis testing image structure. These forms rich information analysing observations statistically robust manner.