Mixed Noise and Posterior Estimation with Conditional DeepGEM

Abstract

Motivated by indirect measurements and applications from nanometrology with a mixed noise model, we develop a novel algorithm for jointly estimating the posterior and the noise parameters in Bayesian inverse problems. We propose to solve the problem by an expectation maximization (EM) algorithm. Based on the current noise parameters, we learn in the E-step a conditional normalizing flow that approximates the posterior. In the M-step, we propose to find the noise parameter updates again by an EM algorithm, which has analytical formulas. We compare the training of the conditional normalizing flow with the forward and reverse KL, and show that our model is able to incorporate information from many measurements, unlike previous approaches.

arXiv commentPublished in Machine Learning: Science and Technology

References in corpus (6)

• Stochastic Normalizing Flows for Inverse Problems: a Markov Chains Viewpoint
• On uncertainties in the reconstruction of nanostructures in EUV scatterometry and grazing incidence small-angle X-ray scattering
• Time-frequency analysis assisted reconstruction of ruthenium optical constants in the sub-EUV spectral range 8nm-23.75nm
• The anisotropy in the optical constants of quartz crystals for soft X-rays
• WPPNets and WPPFlows: The Power of Wasserstein Patch Priors for Superresolution
• Learning Interpolations between Boltzmann Densities