CMB-GAN: Fast Simulations of Cosmic Microwave background anisotropy maps using Deep Learning

Amit Mishra, Pranath Reddy, Rahul NigamarXiv 2019 · arXiv · OpenReview

TL;DR

This paper introduces CMB-GAN, a novel deep learning approach using Generative Adversarial Networks (GANs) to rapidly simulate Cosmic Microwave Background (CMB) anisotropy maps. By drastically reducing computation time from hours to seconds, CMB-GAN offers a powerful new tool for large-scale cosmological experiments that require vast amounts of CMB data.

What Problem Does This Paper Address?

Cosmic Microwave Background (CMB) radiation, the afterglow of the Big Bang, is a treasure trove of information about the universe's origins, composition, and evolution. Cosmologists rely on detailed CMB maps to study fundamental parameters like the universe's geometry, the amount of baryonic matter, and dark energy. However, generating these maps traditionally involves computationally intensive software like CAMB, which can take hours per map. This bottleneck severely limits the scale and speed of cosmological analyses, especially when millions of simulations are needed for parameter estimation and uncertainty quantification. This paper aims to overcome this limitation by developing a much faster method for generating realistic CMB maps.

Methods

The core of this work is the application of Generative Adversarial Networks (GANs) to CMB map generation. The authors train a deep convolutional GAN on 2D patches of CMB maps. The GAN consists of two competing neural networks: a generator that creates synthetic CMB patches and a discriminator that tries to distinguish between real (CAMB-generated) and fake (GAN-generated) patches. Through this adversarial process, the generator learns to produce increasingly realistic CMB data.

To evaluate the quality of the generated maps, the researchers also train a Multilayer Perceptron (MLP) network to predict the baryon density from CMB map patches. This MLP acts as a diagnostic tool: if the GAN-generated maps are sufficiently realistic, the MLP should predict similar baryon densities for both real and GAN-generated patches. Various statistical metrics, including pixel intensity distributions, power spectra, and correlation matrices, are used to quantitatively compare the GAN-generated maps against those produced by CAMB.

Datasets

The primary dataset used for training and evaluation consists of CMB temperature anisotropy maps generated using the standard cosmological code CAMB. The authors generate these maps by inputting various cosmological parameters (e.g., Hubble constant, baryon density, dark matter density). For training the GAN, they extract 2D patches of two sizes: 56° and 112° square regions. These patches are then used to train the GAN and the diagnostic MLP. No new benchmark datasets are introduced; instead, the focus is on generating data that mimics the distribution of existing CAMB simulations.

Key Insights and Ideas

  • GANs as efficient CMB simulators: The paper demonstrates that GANs can learn the complex, non-linear distribution of CMB anisotropy maps and generate new samples orders of magnitude faster than traditional methods.
  • Baryon density as a quality metric: Using an MLP to predict baryon density from CMB patches provides a clever way to assess the realism of GAN-generated maps. If the GAN can fool the MLP into predicting similar baryon densities for real and fake maps, it suggests the GAN has captured crucial cosmological information.
  • Scale invariance: The GAN model shows robustness to the size of the input patches (56° vs. 112°), indicating it can learn underlying CMB characteristics irrespective of the specific patch size.
  • Sensitivity to cosmological parameters: The GAN's output is demonstrably sensitive to changes in input cosmological parameters (specifically, the Hubble constant), suggesting it's not just memorizing data but truly learning the underlying physics.

Key Results

  • Massive speedup: The CMB-GAN model can generate CMB patches in seconds, compared to hours required by CAMB for a single map. This represents an exponential reduction in computational time.
  • High fidelity generation: Statistical comparisons (power spectra, pixel intensity distributions, correlation matrices) show strong agreement between GAN-generated and CAMB-generated CMB patches, indicating the GAN captures essential features of the CMB.
  • Parameter sensitivity: Experiments varying the Hubble constant show that the GAN-generated maps exhibit distinct statistical properties (e.g., standard deviation of pixel intensity, correlation matrices) that correlate with the input parameter, validating the model's ability to reflect cosmological variations.

Limitations and Future Work

  • Patch-based generation: The current work focuses on generating 2D patches rather than full-sky maps. This is acknowledged as a limitation due to computational constraints during training.
  • Spherical convolutions: The authors suggest that using spherical convolutional layers instead of standard 2D convolutions, and training on full-sky CMB patches, could lead to more accurate and comprehensive simulations.
  • Extending to full-sky: The ultimate goal is to extend this methodology to generate full-sky CMB maps, which would be a significant advancement for large-scale cosmological analyses.
  • Further parameter exploration: While the Hubble constant was explored, future work could investigate the GAN's sensitivity to a wider range of cosmological parameters.

Summary updated

Abstract

Cosmic Microwave Background (CMB) has been a cornerstone in many cosmology experiments and studies since it was discovered back in 1964. Traditional computational models like CAMB that are used for generating CMB temperature anisotropy maps are extremely resource intensive and act as a bottleneck in cosmology experiments that require a large amount of CMB data for analysis. In this paper, we present a new approach to the generation of CMB temperature maps using a specific class of neural networks called Generative Adversarial Network (GAN). We train our deep generative model to learn the complex distribution of CMB maps and efficiently generate new sets of CMB data in the form of 2D patches of anisotropy maps without losing much accuracy. We limit our experiment to the generation of 56$^{\circ}$ and 112$^{\circ}$ square patches of CMB maps. We have also trained a Multilayer perceptron model for estimation of baryon density from a CMB map, we will be using this model for the performance evaluation of our generative model using diagnostic measures like Histogram of pixel intensities, the standard deviation of pixel intensity distribution, Power Spectrum, Cross power spectrum, Correlation matrix of the power spectrum and Peak count. We show that the GAN model is able to efficiently generate CMB samples of multiple sizes and is sensitive to the cosmological parameters corresponding to the underlying distribution of the data. The primiary advantage of this method is the exponential reduction in the computational time needed to generate the CMB data, the GAN model is able to generate the samples within seconds as opposed to hours required by the CAMB package with an acceptable value to error and loss of information. We hope that future iterations of this methodology will replace traditional statistical methods of CMB data generation and help in large scale cosmological experiments.
BibTeX
@misc{mishra2019cmb,
  author = {Amit Mishra and Pranath Reddy and Rahul Nigam},
  title = {{CMB-GAN: Fast Simulations of Cosmic Microwave background anisotropy maps using Deep Learning}},
  year = {2019},
  eprint = {1908.04682},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1908.04682},
}
Figure