skip to main content

Title: Cleaning our own dust: simulating and separating galactic dust foregrounds with neural networks
ABSTRACT Separating galactic foreground emission from maps of the cosmic microwave background (CMB) and quantifying the uncertainty in the CMB maps due to errors in foreground separation are important for avoiding biases in scientific conclusions. Our ability to quantify such uncertainty is limited by our lack of a model for the statistical distribution of the foreground emission. Here, we use a deep convolutional generative adversarial network (DCGAN) to create an effective non-Gaussian statistical model for intensity of emission by interstellar dust. For training data we use a set of dust maps inferred from observations by the Planck satellite. A DCGAN is uniquely suited for such unsupervised learning tasks as it can learn to model a complex non-Gaussian distribution directly from examples. We then use these simulations to train a second neural network to estimate the underlying CMB signal from dust-contaminated maps. We discuss other potential uses for the trained DCGAN, and the generalization to polarized emission from both dust and synchrotron.
Authors:
; ; ; ;
Award ID(s):
1812199 1560482 1852581
Publication Date:
NSF-PAR ID:
10253855
Journal Name:
Monthly Notices of the Royal Astronomical Society
Volume:
500
Issue:
3
Page Range or eLocation-ID:
3889 to 3897
ISSN:
0035-8711
Sponsoring Org:
National Science Foundation
More Like this
  1. 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 observational data. In particular, we apply a type of neural network called a variational autoencoder (VAE) to maps of the intensity of emission from interstellar dust as inferred from Planck sky maps and demonstrate its ability to (i) simulate new samples with similar summary statistics as the training set, (ii) provide fits to emission maps withheld from the training set, and (iii) produce constrained realizations. We find VAEs are easier to train than another popular architecture: that of generative adversarial networks, and are better suited for use in Bayesian inference.
  2. Abstract

    We present the first linear polarization measurements from the 2015 long-duration balloon flight ofSpider, which is an experiment that is designed to map the polarization of the cosmic microwave background (CMB) on degree angular scales. The results from these measurements include maps and angular power spectra from observations of 4.8% of the sky at 95 and 150 GHz, along with the results of internal consistency tests on these data. While the polarized CMB anisotropy from primordial density perturbations is the dominant signal in this region of sky, Galactic dust emission is also detected with high significance. Galactic synchrotron emission is found to be negligible in theSpiderbands. We employ two independent foreground-removal techniques to explore the sensitivity of the cosmological result to the assumptions made by each. The primary method uses a dust template derived fromPlanckdata to subtract the Galactic dust signal. A second approach, which constitutes a joint analysis ofSpiderandPlanckdata in the harmonic domain, assumes a modified-blackbody model for the spectral energy distribution of the dust with no constraint on its spatial morphology. Using a likelihood that jointly samples the template amplitude andrparameter space, we derive 95% upper limits on the primordial tensor-to-scalar ratio from Feldman–Cousins and Bayesian constructions,more »findingr< 0.11 andr< 0.19, respectively. Roughly half the uncertainty inrderives from noise associated with the template subtraction. New data at 280 GHz fromSpider’s second flight will complement thePlanckpolarization maps, providing powerful measurements of the polarized Galactic dust emission.

    « less
  3. ABSTRACT

    In order to better analyse the polarization of the cosmic microwave background (CMB), which is dominated by emission from our Galaxy, we need tools that can detect residual foregrounds in cleaned CMB maps. Galactic foregrounds introduce statistical anisotropy and directionality to the polarization pseudo-vectors of the CMB, which can be investigated by using the $\mathcal {D}$ statistic of Bunn and Scott. This statistic is rapidly computable and capable of investigating a broad range of data products for directionality. We demonstrate the application of this statistic to detecting foregrounds in polarization maps by analysing the uncleaned Planck 2018 frequency maps. For the Planck 2018 CMB maps, we find no evidence for residual foreground contamination. In order to examine the sensitivity of the $\mathcal {D}$ statistic, we add a varying fraction of the polarized thermal dust and synchrotron foreground maps to the CMB maps and show the per cent-level foreground contamination that would be detected with 95 per cent confidence. We also demonstrate application of the $\mathcal {D}$ statistic to another data product by analysing the gradient of the minimum-variance CMB lensing potential map (i.e. the deflection angle) for directionality. We find no excess directionality in the lensing potential map when compared to the simulationsmore »provided by the Planck Collaboration.

    « less
  4. For the past decade, the BICEP/Keck collaboration has been operating a series of telescopes at the Amundsen-Scott South Pole Station measuring degree-scale B-mode polarization imprinted in the Cosmic Microwave Background (CMB) by primordial gravitational waves (PGWs). These telescopes are compact refracting polarimeters mapping about 2% of the sky, observing at a broad range of frequencies to account for the polarized foreground from Galactic synchrotron and thermal dust emission. Our latest publication "BK18" utilizes the data collected up to the 2018 observing season, in conjunction with the publicly available WMAP and Planck data, to constrain the tensor-to-scalar ratio r. It particularly includes (1) the 3-year BICEP3 data which is the current deepest CMB polarization map at the foreground-minimum 95 GHz; and (2) the Keck 220 GHz map with a higher signal-to-noise ratio on the dust foreground than the Planck 353 GHz map. We fit the auto- and cross-spectra of these maps to a multicomponent likelihood model (ΛCDM+dust+synchrotron+noise+r) and find it to be an adequate description of the data at the current noise level. The likelihood analysis yields σ(r)=0.009. The inference of r from our baseline model is tightened to r0.05=0.014+0.010−0.011 and r0.05<0.036 at 95% confidence, meaning that the BICEP/Keck B-mode datamore »is the most powerful existing dataset for the constraint of PGWs. The up-coming BICEP Array telescope is projected to reach σ(r)≲0.003 using data up to 2027.« less
  5. ABSTRACT We present a novel technique for cosmic microwave background (CMB) foreground subtraction based on the framework of blind source separation. Inspired by previous work incorporating local variation to generalized morphological component analysis (GMCA), we introduce hierarchical GMCA (HGMCA), a Bayesian hierarchical graphical model for source separation. We test our method on Nside = 256 simulated sky maps that include dust, synchrotron, free–free, and anomalous microwave emission, and show that HGMCA reduces foreground contamination by $25{{\ \rm per\ cent}}$ over GMCA in both the regions included and excluded by the Planck UT78 mask, decreases the error in the measurement of the CMB temperature power spectrum to the 0.02–0.03 per cent level at ℓ > 200 (and $\lt 0.26{{\ \rm per\ cent}}$ for all ℓ), and reduces correlation to all the foregrounds. We find equivalent or improved performance when compared to state-of-the-art internal linear combination type algorithms on these simulations, suggesting that HGMCA may be a competitive alternative to foreground separation techniques previously applied to observed CMB data. Additionally, we show that our performance does not suffer when we perturb model parameters or alter the CMB realization, which suggests that our algorithm generalizes well beyond our simplified simulations. Our results open a new avenuemore »for constructing CMB maps through Bayesian hierarchical analysis.« less