We present an upgraded combined estimator, based on Minkowski Functionals and Neural Networks, with excellent performance in detecting primordial non-Gaussianity in simulated maps that also contain a weighted mixture of Galactic contaminations, besid
es real pixels noise from Planck cosmic microwave background radiation data. We rigorously test the efficiency of our estimator considering several plausible scenarios for residual non-Gaussianities in the foreground-cleaned Planck maps, with the intuition to optimize the training procedure of the Neural Network to discriminate between contaminations with primordial and secondary non-Gaussian signatures. We look for constraints of primordial local non-Gaussianity at large angular scales in the foreground-cleaned Planck maps. For the $mathtt{SMICA}$ map we found ${f}_{rm ,NL} = 33 pm 23$, at $1sigma$ confidence level, in excellent agreement with the WMAP-9yr and Planck results. In addition, for the other three Planck maps we obtain similar constraints with values in the interval ${f}_{rm ,NL} in [33, 41]$, concomitant with the fact that these maps manifest distinct features in reported analyses, like having different pixels noise intensities.
The extensive search for deviations from Gaussianity in cosmic microwave background radiation (CMB) data is very important due to the information about the very early moments of the universe encoded there. Recent analyses from Planck CMB data do not
exclude the presence of non-Gaussianity of small amplitude, although they are consistent with the Gaussian hypothesis. The use of different techniques is essential to provide information about types and amplitudes of non-Gaussianities in the CMB data. In particular, we find interesting to construct an estimator based upon the combination of two powerful statistical tools that appears to be sensitive enough to detect tiny deviations from Gaussianity in CMB maps. This estimator combines the Minkowski functionals with a Neural Network, maximizing a tool widely used to study non-Gaussian signals with a reinforcement of another tool designed to identify patterns in a data set. We test our estimator by analyzing simulated CMB maps contaminated with different amounts of local primordial non-Gaussianity quantified by the dimensionless parameter fNL. We apply it to these sets of CMB maps and find gtrsim 98% of chance of positive detection, even for small intensity local non-Gaussianity like fNL = 38 +/- 18, the current limit from Planck data for large angular scales. Additionally, we test the suitability to distinguish between primary and secondary non-Gaussianities and find out that our method successfully classifies ~ 95% of the tested maps. Furthermore, we analyze the foreground-cleaned Planck maps obtaining constraints for non-Gaussianity at large-angles that are in good agreement with recent constraints. Finally, we also test the robustness of our estimator including cut-sky masks and realistic noise maps measured by Planck, obtaining successful results as well.
