Hybrid Physical-Deep Learning Model for Astronomical Inverse Problems
This work addresses inverse problems in astronomy by providing a method that combines physical knowledge with data-driven priors, though it appears incremental as it builds on existing hybrid approaches.
The authors tackled the problem of astronomical source separation by developing a hybrid Bayesian machine learning architecture that integrates physical parametrization with a deep generative prior, resulting in an interpretable and differentiable model applicable to varying data qualities without retraining.
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.