I will present a Bayesian machine learning architecture that combines a physically motivated parameterization 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. I will demonstrate this approach with an example of astronomical source separation in current imaging data, yielding a physical and interpretable model of astronomical scenes.
Princeton University Machine Learning for the Sciences
Solving Inverse Problems with Data-driven Priors
Lunch will be served.
Date & Time
January 17, 2020 | 12:00 – 1:00pm
Center for Statistics and Machine Learning (CSML) 26 Prospect Ave Auditorium 103