Identifying Cosmological Information in a Deep Neural Network

TL;DR

This paper introduces a convolutional neural network approach to estimate dark matter particle mass from cosmological images, outperforming traditional statistical methods and demonstrating the potential of deep learning in cosmology.

Contribution

It presents a novel CNN-based method for constraining dark matter mass from images, showing improved performance over traditional two-point correlation functions.

Findings

CNN outperforms two-point correlation in dark matter mass estimation

Random Gaussian trained CNN performs comparably to traditional methods

Deep learning offers a promising complementary tool for cosmological parameter inference

Abstract

A novel method images to estimate cosmological parameters based on images is presented. In this paper, we demonstrate the use of a convolutional neural network (CNN) for constraining the mass of dark matter particle. For this purpose, we perform a suite of N-body simulations with different dark matter particle masses to train CNN and estimate dark matter mass using a density-contrast map. The proposed method is complementary to the one based on summary statistics, such as two-point correlation function. We compare our CNN classification results with those obtained from the two-point correlation of the distribution of dark matter particles, and find that the CNN offers better performance In addition, we use images made from a random Gauss simulation to train a CNN, which is then compared with the CNN trained by N-body simulation and two-point correlation. The random Gauss-trained CNN has comparable performance to two-point correlation.