Markov Chain Monte Carlo (MCMC) algorithms are commonly used for their versatility in sampling from complicated probability distributions. However, as the dimension of the distribution gets larger, the computational costs for a satisfactory exploration of the sampling space become challenging. Adaptive MCMC methods employing a choice of proposal distribution can address this issue speeding up the convergence. In this paper we show an alternative way of performing adaptive MCMC, by using the outcome of Bayesian Neural Networks as the initial proposal for the Markov Chain. This combined approach increases the acceptance rate in the Metropolis-Hasting algorithm and accelerate the convergence of the MCMC while reaching the same final accuracy. Finally, we demonstrate the main advantages of this approach by constraining the cosmological parameters directly from Cosmic Microwave Background maps.

In this paper, we present the first study that compares different models of Bayesian neural networks (BNNs) to predict the posterior distribution of the cosmological parameters directly from the cosmic microwave background (CMB) temperature and polarization maps. We focus our analysis on four different methods to sample the weights of the network during training: Dropout, DropConnect, Reparameterization Trick (RT), and Flipout. We find that Flipout outperforms all other methods regardless of the architecture used, and provides tighter constraints for the cosmological parameters. Moreover, we compare our results with a Markov chain Monte Carlo (MCMC) posterior analysis and obtain comparable error correlations among parameters, with BNNs that are orders of magnitude faster in inference, albeit less accurate. Thanks to the speed of the inference process with BNNs, the posterior distribution—the outcome of the neural network—can be used as the initial proposal for the Markov chain. We show that this combined approach increases the acceptance rate in the Metropolis-Hasting algorithm and accelerates the convergence of the MCMC, while reaching the same final accuracy. In the second part of the paper, we present a guide to the training and calibration of a successful multichannel BNN for the CMB temperature and polarization map. We show how tuning the regularization parameter for the standard deviation of the approximate posterior on the weights in Flipout and RT can produce unbiased and reliable uncertainty estimates, i.e., the regularizer acts like a hyperparameter analogous to the dropout rate in Dropout. The best performances are nevertheless achieved with a more convenient method, in which the network parameters are kept free during training to achieve the best uncalibrated performances, and the confidence intervals are calibrated in a subsequent phase. Additionally, we describe existing strategies for calibrating the networks and propose new ones. Finally, we show how polarization, when combined with the temperature in a unique multichannel tensor fed to a single BNN, helps to break degeneracies among parameters and provides stringent constraints. The results reported in this paper can be extended to other cosmological data sets in order to capture features that can be extracted directly from the raw data, such as non-Gaussianity or foreground emissions.

Upcoming experiments such as Hydrogen Epoch of Reionization Array(HERA) and the Square Kilometre Array (SKA) are intended to measure the 21 cm signal over a wide range of redshifts, representing an incredible opportunity in advancing our understanding about the nature of cosmic reionization. At the same time these kind of experiments will present new challenges in processing the extensive amount of data generated, calling for the development of automated methods capable of precisely estimating physical parameters and their uncertainties. In this deliverable we employ Variational Inference, and in particular Bayesian Neural Networks, as an alternative to MCMC in 21 cm observations to report credible estimations for cosmological and astrophysical parameters and assess the correlations among them. Finally, we have implemented the use of bijectors to improve the diagonal Gaussian approximate posteriors and be able to extract significant information from Non-Gaussian signal in the 21 cm dataset.

Upcoming experiments such as Hydrogen Epoch of Reionization Array (HERA) and Square Kilometre Array (SKA) are intended to measure the 21cm signal over a wide range of redshifts, representing an incredible opportunity in advancing our understanding about the nature of cosmic Reionization. At the same time these kind of experiments will present new challenges in processing the extensive amount of data generated, calling for the development of automated methods capable of precisely estimating physical parameters and their uncertainties. In this paper we employ Variational Inference, and in particular Bayesian Neural Networks, as an alternative to MCMC in 21 cm observations to report credible estimations for cosmological and astrophysical parameters and assess the correlations among them.

Bayesian Neural Networks (BNNs) often result uncalibrated after training, usually tending towards overconfidence. Devising effective calibration methods with low impact in terms of computational complexity is thus of central interest. In this paper we present calibration methods for BNNs based on the alpha divergences from Information Geometry. We compare the use of alpha divergence in training and in calibration, and we show how the use in calibration provides better calibrated uncertainty estimates for specific choices of alpha and is more efficient especially for complex network architectures. We empirically demonstrate the advantages of alpha calibration in regression problems involving parameter estimation and inferred correlations between output uncertainties.

In this paper, we investigate the effects of helical primordial magnetic fields (PMFs) on the cosmic microwave background (CMB) reduced bispectrum. We derive the full three-point statistics of helical magnetic fields and numerically calculate the even contribution in the collinear configuration. We then numerically compute the CMB reduced bispectrum induced by passive and compensated PMF modes on large angular scales. There is a negative signal on the bispectrum due to the helical terms of the fields and we also observe that the biggest contribution to the bispectrum comes from the non-zero IR cut-off for causal fields, unlike the two-point correlation case. For negative spectral indices, the reduced bispectrum is enhanced by the passive modes. This gives a lower value of the upper limit for the mean amplitude of the magnetic field on a given characteristic scale. However, high values of IR cut-off in the bispectrum, and the helical terms of the magnetic field relaxes this bound. This demonstrates the importance of the IR cut-off and helicity in the study of the nature of PMFs from CMB observations.

In this paper we contrasted two cosmological perturbation theory formalisms, the 1+3 covariant gauge invariant and the gauge invariant by comparing their gauge invariant variables associated with magnetic field defined in each approach. In the first part we give an introduction to each formalism assuming the presence of a magnetic field. We found that gauge invariant quantities defined by 1+3 covariant approach are related with spatial variations of the magnetic field (defined in the gauge invariant formalism) between two closed fundamental observers. This relation was computed by choosing the comoving gauge in the gauge invariant approach in a magnetized universe. Furthermore, we have derived the gauge transformations for electromagnetic potentials in the gauge invariant approach and the Maxwell's equations have been written in terms of these potentials.

The origin of large scale magnetic fields is one of the most puzzling topics in cosmology and astrophysics. It is assumed that the observed magnetic fields result from the amplification of an initial field produced in the early universe. In this paper we compute the exact power spectrum of magnetic fields created after inflation best known as post inflationary magnetic fields, using the first order cosmological perturbation theory. Our treatment differs from others works because we include an infrared cutoff which encodes only causal modes in the spectrum. The cross-correlation between magnetic energy density with Lorentz force and the anisotropic part of the electromagnetic field are exactly computed. We compare our results with previous works finding agreement in cases where the ratio between lower and upper cutoff is very small. However, we found that spectrum is strongly affected when this ratio is greater than 0.2. Moreover, the effect of a post inflationary magnetic field with a lower cutoff on the angular power spectrum in the temperature distribution of CMB was also exactly calculated. The main feature is a shift of the spectrum's peak as function of the infrared cutoff, therefore analyzing this effect we could infer the value of this cutoff and thus constraining the primordial magnetic fields generation models.

The origin of galactic and extra-galactic magnetic fields is an unsolved problem in modern cosmology. A possible scenario comes from the idea of these fields emerged from a small field, a seed, which was produced in the early universe (phase transitions, inflation, ...) and it evolves in time. Cosmological perturbation theory offers a natural way to study the evolution of primordial magnetic fields. The dynamics for this field in the cosmological context is described by a cosmic dynamo like equation, through the dynamo term. In this paper we get the perturbed Maxwell's equations and compute the energy momentum tensor to second order in perturbation theory in terms of gauge invariant quantities. Two possible scenarios are discussed, first we consider a FLRW background without magnetic field and we study the perturbation theory introducing the magnetic field as a perturbation. The second scenario, we consider a magnetized FLRW and build up the perturbation theory from this background. We compare the cosmological dynamo like equation in both scenarios

Argo is a library for deep learning algorithms based on TensorFlow and Sonnet. The library allows you to train different models (feed-forwards neural networks for regression and classification problems, autoencoders and variational autoencoders, Bayesian neural networks, Helmholtz machines, etc) by specifying their parameters as well as the network topologies in a configuration file. The models can then be trained in parallel in presence of multiple GPUs. The library is easy to expand for alternative models and training algorithms, as well as for different network topologies.