Improving Convolutional Neural Networks for Cosmological Fields with Random Permutation

TL;DR

This paper introduces a novel regularization and data augmentation technique involving pixel shuffling in CNNs, significantly improving parameter inference accuracy in cosmological mass maps, especially under low signal-to-noise conditions.

Contribution

The authors propose a simple yet effective regularization method using pixel permutation and shuffling, enhancing CNN performance for cosmological map analysis beyond existing methods.

Findings

30% improvement in $S_8$ parameter constraints for simulated surveys

Robust performance gains in low signal-to-noise regimes

Potential applicability to other cosmological data types

Abstract

Convolutional Neural Networks (CNNs) have recently been applied to cosmological fields -- weak lensing mass maps and galaxy maps. However, cosmological maps differ in several ways from the vast majority of images that CNNs have been tested on: they are stochastic, typically low signal-to-noise per pixel, and with correlations on all scales. Further, the cosmology goal is a regression problem aimed at inferring posteriors on parameters that must be unbiased. We explore simple CNN architectures and present a novel approach of regularization and data augmentation to improve its performance for lensing mass maps. We find robust improvement by using a mixture of pooling and shuffling of the pixels in the deep layers. The random permutation regularizes the network in the low signal-to-noise regime and effectively augments the existing data. We use simulation-based inference (SBI) to show that the model outperforms CNN designs in the literature. We find a 30% improvement in the constraints of the $S_8$ parameter for simulated Stage-III surveys, including systematic uncertainties such as intrinsic alignments. We explore various statistical errors corresponding to next-generation surveys and find comparable improvements. We expect that our approach will have applications to other cosmological fields as well, such as galaxy maps or 21-cm maps.