A Generative Model of Galactic Dust Emission Using Variational Inference

TL;DR

This paper introduces a variational autoencoder approach to model galactic dust emission, aiding CMB observation planning and foreground cleaning by capturing statistical properties from observational data and simulations.

Contribution

It demonstrates the effectiveness of VAEs in simulating, fitting, and producing constrained realizations of dust emission maps, offering advantages over GANs for Bayesian inference.

Findings

VAEs can generate realistic dust emission samples.

VAEs outperform GANs in training stability and Bayesian applications.

The method aids in understanding and modeling galactic dust for CMB studies.

Abstract

Emission from the interstellar medium can be a significant contaminant of measurements of the intensity and polarization of the cosmic microwave background (CMB). For planning CMB observations, and for optimizing foreground-cleaning algorithms, a description of the statistical properties of such emission can be helpful. Here we examine a machine learning approach to inferring the statistical properties of dust from either observational data or physics-based simulations. In particular, we apply a type of neural network called a Variational Auto Encoder (VAE) to maps of the intensity of emission from interstellar dust as inferred from Planck sky maps and demonstrate its ability to a) simulate new samples with similar summary statistics as the training set, b) provide fits to emission maps withheld from the training set, and c) produce constrained realizations. We find VAEs are easier to train than another popular architecture: that of Generative Adversarial Networks (GANs), and are better-suited for use in Bayesian inference.