Encoding large-scale cosmological structure with generative adversarial networks

Abstract

Recently, a type of neural networks called generative adversarial networks (GANs) has been proposed as a solution for the fast generation of simulation-like datasets in an attempt to avoid intensive computations and running cosmological simulations that are expensive in terms of time and computing power. We built and trained a GAN to determine the strengths and limitations of such an approach in more detail. We then show how we made use of the trained GAN to construct an autoencoder (AE) that can conserve the statistical properties of the data. The GAN and AE were trained on images and cubes issued from two types of N-body simulations, namely 2D and 3D simulations. We find that the GAN successfully generates new images and cubes that are statistically consistent with the data on which it was trained. We then show that the AE can efficiently extract information from simulation data and satisfactorily infers the latent encoding of the GAN to generate data with similar large-scale structures.