Bayesian decomposition of the Galactic multi-frequency sky using probabilistic autoencoders

Abstract

All-sky observations of the Milky Way show both Galactic and non-Galactic diffuse emission, for example from interstellar matter or the cosmic microwave background (CMB). The different emitters are partly superimposed in the measurements, partly they obscure each other, and sometimes they dominate within a certain spectral range. The decomposition of the underlying radiative components from spectral data is a signal reconstruction problem and often associated with detailed physical modeling and substantial computational effort. We aim to build an effective and self-instructing algorithm detecting the essential spectral information contained Galactic all-sky data covering spectral bands from $\gamma$-ray to radio waves. Utilizing principles from information theory, we develop a state-of-the-art variational autoencoder specialized on the adaption to Gaussian noise statistics. We first derive a generic generative process that leads from a low-dimensional set of emission features to the observed high-dimensional data. We formulate a posterior distribution of these features using Bayesian methods and approximate this posterior with variational inference. The algorithm efficiently encodes the information of 35 Galactic emission data sets in ten latent feature maps. These contain the essential information required to reconstruct the initial data with high fidelity and are ranked by the algorithm according to their significance for data regeneration. The three most significant feature maps encode astrophysical components: (1) The dense interstellar medium (ISM), (2) the hot and dilute regions of the ISM and (3) the CMB. The machine-assisted and data-driven dimensionality reduction of spectral data is able to uncover the physical features encoding the input data. Our algorithm is able to extract the dense and dilute Galactic regions, as well as the CMB, from the sky brightness values only.

arXiv comment25 pages, 8 figures, 3 tables. Submitted to Astronomy & Astrophysics

References in corpus (30)

• HEALPix -- a Framework for High Resolution Discretization, and Fast Analysis of Data Distributed on the Sphere
• The Large Area Telescope on the Fermi Gamma-ray Space Telescope Mission
• Variational Inference: A Review for Statisticians
• Star Formation in the Milky Way and Nearby Galaxies
• Planck 2018 results. I. Overview and the cosmological legacy of Planck
• HI4PI: A full-sky HI survey based on EBHIS and GASS
• The Interstellar Environment of our Galaxy
• A Full-Sky H-alpha Template for Microwave Foreground Prediction
• The Fermi Large Area Telescope On Orbit: Event Classification, Instrument Response Functions, and Calibration
• IRIS: A new generation of IRAS maps
• A Robotic Wide-Angle H-Alpha Survey of the Southern Sky
• Seven-Year Wilkinson Microwave Anisotropy Probe (WMAP) Observations: Galactic Foreground Emission
• Planck 2018 results. IV. Diffuse component separation
• Revised Catalog of GALEX Ultraviolet Sources. I. The All-sky Survey: GUVcat_AIS
• Joint Bayesian component separation and CMB power spectrum estimation
• Planck 2015 results. X. Diffuse component separation: Foreground maps
• Component separation methods for the Planck mission
• Multi-Detector Multi-Component spectral matching and applications for CMB data analysis
• Surveying the reach and maturity of machine learning and artificial intelligence in astronomy
• Big Universe, Big Data: Machine Learning and Image Analysis for Astronomy
• The $AKARI$ Far-Infrared All-Sky Survey Maps
• Dimensionality Reduction of SDSS Spectra with Variational Autoencoders
• The Denoised, Deconvolved, and Decomposed Fermi $\gamma$-ray sky - An application of the D${}^3$PO algorithm
• Information theory for fields
• Classifying structures in the ISM with Support Vector Machines: the G16.05-0.57 supernova remnant
• GAME: GAlaxy Machine learning for Emission lines
• Foreword to the Focus Issue on Machine Learning in Astronomy and Astrophysics
• Optimal Belief Approximation
• The interstellar medium of dwarf galaxies: new insights from Machine Learning analysis of emission line spectra
• Sharpening up Galactic all-sky maps with complementary data - A machine learning approach

Cited by in corpus (3)

• Using autoencoders and deep transfer learning to determine the stellar parameters of 286 CARMENES M dwarfs
• Quantifying imperfect cognition via achieved information gain
• Probabilistic Autoencoder using Fisher Information