Simulation-based inference with scattering representations: scattering is all you need
This work addresses simulation-based inference for researchers in cosmology and related fields, offering an interpretable and resilient method, though it is incremental as it builds on existing scattering representation techniques.
The paper tackled the problem of simulation-based inference with images by using scattering representations without further compression, demonstrating in a cosmological case study that this approach extracts more information than traditional second-order summary statistics.
We demonstrate the successful use of scattering representations without further compression for simulation-based inference (SBI) with images (i.e. field-level), illustrated with a cosmological case study. Scattering representations provide a highly effective representational space for subsequent learning tasks, although the higher dimensional compressed space introduces challenges. We overcome these through spatial averaging, coupled with more expressive density estimators. Compared to alternative methods, such an approach does not require additional simulations for either training or computing derivatives, is interpretable, and resilient to covariate shift. As expected, we show that a scattering only approach extracts more information than traditional second order summary statistics.