Mixture of noises and sampling of non-log-concave posterior distributions

International audienceThis work considers a challenging radio-astronomyinverse problem of physical parameter inference from multispec-tral observations. The forward model underlying this problem isa computationally expensive numerical simulation. In addition,the observation model mixes different sources of noise yieldinga non-concave log-likelihood function. To overcome these issues,we introduce a likelihood approximation with controlled error.Given the absence of ground truth, parameter inference isconducted with a Markov chain Monte Carlo (MCMC) algorithmto provide credibility intervals along with point estimates. To thisaim, we propose a new sampler that addresses the numericalchallenges induced by the observation model, in particular thenon-log-concavity of the posterior distribution. The efficiency ofthe proposed method is demonstrated on synthetic yet realisticastrophysical data. We believe that the proposed approach isvery general and can be adapted to many similar difficult inverseproblem