Day 1492

Monday.

1810.11027
On the dissection of degenerate cosmologies with machine learning
Merten, Giocoli, et al

Based on the DUSTGRAIN-pathfinder suite of simulations, investigate observational degeneracies between 9 models of modified gravity and massive neutrinos.  3 types of machine learning techniques are tested for their ability to discriminate lensing convergence maps by extracting dimensional reduced representation of the data.  Classical map descriptors such as the power spectrum, peak counts and Minkowski functional are combined into a joint feature vector and compared to the descriptors and statistics that are common to the field of digital image processing.  To learn new features directly from the data, use a Convolutional Neural Network (CNN).  For the mapping between feature vectors and the predictions of their underlying model, implement 2 different classifiers; one based on a nearest-neighbour search and one that is based on a fully connected neural network.  Find that the neural network provides a much more robust classification than the nearest-neighbor approach and that the CNN provides the most discriminating representation of the data.  It achieves the cleanest separation between the different models and the highest classification success rate of 59% for a single source redshift.  Once the tomographic CNN analysis is preformed, the total classification accuracy increases significantly to 76% with no observational degeneracies remaining.  Visualizing the filter response of the CNN at different network depths provides us with the unique opportunity to learn from very complex models and to understand better why they perform so well.


1810.11040
The effect of dark matter-dark radiation interactions on halo abundance -- a Press-Schechter approach
Sameie, et al

Study halo mass functions with the PS formalism for interaction DM models, where matter PS are damped due to dark acoustic oscillations in the early universe.  After adopting a smooth window function, calibrate the analytical model with numerical simulations from the "Effective theory of structure formation" (ETHOS) project and fix the model parameters in the high mass regime, M_h >~ 3e10 Msun.  Also perform high- resolution cosmological simulations with halo masses down to M_h~1e8 Msun to cover a wide mass range for comparison.  Although the model is calibrated with ETHOS1 and CDM simulations for high halo masses at z=0, it successfully reproduces simulations for 2 other ETHOS models in the low mass regime at low and high redshifts.  As an application, compare the cumulative number density of haloes to that of observed galaxies at z=6, and find the interacting DM models with a kinetic decoupling temperature below 0.5 keV is favored.  Also perform the abundance-matching analysis and derive the stellar-halo mass relation for these models at z=4.  Suppression in halo abundance leads to less massive haloes that host observed galaxies in the stellar mass range M*~=1e5-1e7 Msun.