Abstract
We demonstrate the successful use of scattering representations withoutfurther compression for simulation-based inference (SBI) with images (i.e.field-level), illustrated with a cosmological case study. Scatteringrepresentations provide a highly effective representational space forsubsequent learning tasks, although the higher dimensional compressed spaceintroduces challenges. We overcome these through spatial averaging, coupledwith more expressive density estimators. Compared to alternative methods, suchan approach does not require additional simulations for either training orcomputing derivatives, is interpretable, and resilient to covariate shift. Asexpected, we show that a scattering only approach extracts more informationthan traditional second order summary statistics.