We present a Bayesian machine learning architecture that combines a physically motivated parametrization and an analytic error model for the likelihood with a deep generative model providing a powerful data-driven prior for complex signals. This combination yields an interpretable and differentiable generative model, allows the incorporation of prior knowledge, and can be utilized for observations with different data quality without having to retrain the deep network. We demonstrate our approach with an example of astronomical source separation in current imaging data, yielding a physical and interpretable model of astronomical scenes.
@article{arxiv.1912.03980,
title = {Hybrid Physical-Deep Learning Model for Astronomical Inverse Problems},
author = {Francois Lanusse and Peter Melchior and Fred Moolekamp},
journal= {arXiv preprint arXiv:1912.03980},
year = {2019}
}
Comments
8 pages, accepted submission to the NeurIPS 2019 Machine Learning and the Physical Sciences Workshop