We propose a novel approach for mitigating radio frequency interference (RFI) signals in radio data using the latest advances in deep learning. We employ a special type of Convolutional Neural Network, the U-Net, that enables the classification of clean signal and RFI signatures in 2D time-ordered data acquired from a radio telescope. We train and assess the performance of this network using the HIDE & SEEK radio data simulation and processing packages, as well as early Science Verification data acquired with the 7m single-dish telescope at the Bleien Observatory. We find that our U-Net implementation is showing competitive accuracy to classical RFI mitigation algorithms such as SEEK's SumThreshold implementation. We publish our U-Net software package on GitHub under GPLv3 license.

To shed light on the fundamental problems posed by dark energy and dark matter, a large number of experiments have been performed and combined to constrain cosmological models. We propose a novel way of quantifying the information gained by updates on the parameter constraints from a series of experiments which can either complement earlier measurements or replace them. For this purpose, we use the Kullback-Leibler divergence or relative entropy from information theory to measure differences in the posterior distributions in model parameter space from a pair of experiments. We apply this formalism to a historical series of cosmic microwave background experiments ranging from Boomerang to WMAP, SPT, and Planck. Considering different combinations of these experiments, we thus estimate the information gain in units of bits and distinguish contributions from the reduction of statistical errors and the "surprise" corresponding to a significant shift of the parameters' central values. For this experiment series, we find individual relative entropy gains ranging from about 1 to 30 bits. In some cases, e.g. when comparing WMAP and Planck results, we find that the gains are dominated by the surprise rather than by improvements in statistical precision. We discuss how this technique provides a useful tool for both quantifying the constraining power of data from cosmological probes and detecting the tensions between experiments.

Abstract. Bayesian inference is often used in cosmology and astrophysics to derive constraints on model parameters from observations. This approach relies on the ability to compute the likelihood of the data given a choice of model parameters. In many practical situations, the likelihood function may however be unavailable or intractable due to non-gaussian errors, non-linear measurements processes, or complex data formats such as catalogs and maps. In these cases, the simulation of mock data sets can often be made through forward modeling. We discuss how Approximate Bayesian Computation (ABC) can be used in these cases to derive an approximation to the posterior constraints using simulated data sets. This technique relies on the sampling of the parameter set, a distance metric to quantify the difference between the observation and the simulations and summary statistics to compress the information in the data. We first review the principles of ABC and discuss its implementation using a Population Monte-Carlo (PMC) algorithm and the Mahalanobis distance metric. We test the performance of the implementation using a Gaussian toy model. We then apply the ABC technique to the practical case of the calibration of image simulations for wide field cosmological surveys. We find that the ABC analysis is able to provide reliable parameter constraints for this problem and is therefore a promising technique for other applications in cosmology and astrophysics. Our implementation of the ABC PMC method is made available via a public code release.

We study the benefits and limits of parallelised Markov chain Monte Carlo (MCMC) sampling in cosmology. MCMC methods are widely used for the estimation of cosmological parameters from a given set of observations and are typically based on the Metropolis-Hastings algorithm. Some of the required calculations can however be computationally intensive, meaning that a single long chain can take several hours or days to calculate. In practice, this can be limiting, since the MCMC process needs to be performed many times to test the impact of possible systematics and to understand the robustness of the measurements being made. To achieve greater speed through parallelisation, MCMC algorithms need to have short auto-correlation times and minimal overheads caused by tuning and burn-in. The resulting scalability is hence influenced by two factors, the MCMC overheads and the parallelisation costs. In order to efficiently distribute the MCMC sampling over thousands of cores on modern cloud computing infrastructure, we developed a Python framework called CosmoHammer which embeds emcee, an implementation by Foreman-Mackey et al. (2012) of the affine invariant ensemble sampler by Goodman and Weare (2010). We test the performance of CosmoHammer for cosmological parameter estimation from cosmic microwave background data. While Metropolis-Hastings is dominated by overheads, CosmoHammer is able to accelerate the sampling process from a wall time of 30 hours on a dual core notebook to 16 minutes by scaling out to 2048 cores. Such short wall times for complex data sets opens possibilities for extensive model testing and control of systematics.Comment: Published version. 17 pages, 6 figures. The code is available at http://www.astro.ethz.ch/refregier/research/Software/cosmohamme

Intensity mapping of neutral hydrogen (HI) is a promising observational probe of cosmology and large-scale structure. We present wide field simulations of HI intensity maps based on N-body simulations of a 2.6 Gpc/h box with 2048 3 particles (particle mass 1.6 × 10 11 M /h). Using a conditional mass function to populate the simulated dark matter density field with halos below the mass resolution of the simulation (10 8 M /h < M halo < 10 13 M /h), we assign HI to those halos according to a phenomenological halo to HI mass relation. The simulations span a redshift range of 0.35 z 0.9 in redshift bins of width ∆z ≈ 0.05 and cover a quarter of the sky at an angular resolution of about 7 . We use the simulated intensity maps to study the impact of non-linear effects and redshift space distortions on the angular clustering of HI. Focusing on the autocorrelations of the maps, we apply and compare several estimators for the angular power spectrum and its covariance. We verify that these estimators agree with analytic predictions on large scales and study the validity of approximations based on Gaussian random fields, particularly in the context of the covariance. We discuss how our results and the simulated maps can be useful for planning and interpreting future HI intensity mapping surveys.ArXiv ePrint: 1509.01589 arXiv:1509.01589v2 [astro-ph.CO] 1 Mar 2016 1 We do note that recent measurements of the clustering of DLA systems at redshifts 2 z 3 [42] are discrepant with direct HI observations on which our model is built [21]. The resolution of this discrepancy is unclear at present, and we will return to this issue in future work.