Presentation #230.01 in the session “Cosmology”.
Normalizing flows provide a powerful and efficient method for learning a complex, high-dimensional probability distribution from data. This allows one to calculate posteriors, and to sample from the data distribution for forward modeling and data augmentation. For these purposes, we have developed PZFlow, a GPU enabled python package for out-of-the-box modeling of tabular data in astrophysics and cosmology. We will demonstrate PZFlow as a generative model for the LSST DESC RAIL project, which aims to forward model the impact of various systematic errors on photo-z estimation.