Matthew Craigie - Searching for Parity Violation in the Galaxy Distribution with Deep Learning

Abstract: Detecting parity violation in the galaxy distribution traditionally relies on measuring higher-order correlation functions and accurately modeling their noise properties through mock catalogs - a computationally intensive approach that is sensitive to simulation choices. We present a deep learning method that can detect parity violation directly from data without requiring mock catalogs. We demonstrate that this framework can successfully measure the parity violation in the galaxy distribution in a statistically robust way. We also introduce a new model architecture called the Neural Field Scattering Transform (NFST), which enhances the Wavelet Scattering Transform by incorporating trainable neural field filters. Using a simplified dataset, we demonstrate that the NFST can detect parity violation with up to 32 times less data than conventional methods while remaining interpretable. This approach offers a promising new direction for detecting cosmological parity violation in upcoming surveys like DESI, complementing existing approaches by reducing dependence on simulations.