17. December 2021 - 11:30
Tony Bonnaire (Universite Paris-Saclay)
Abstract: The spatial distribution of matter depicts a complex pattern commonly referred to as the cosmic web exhibiting massive nodes that are linked together by uni-dimensional bridges of matter, the filaments, themselves found at the intersection between thin and mildly-dense planes of matter labelled walls forming the vast shells enclosing underdense volumes almost devoided of galaxies called voids. First taking a step back from cosmological questions, we propose several algorithms aiming at learning patterns in point-cloud datasets such as a sparse distribution of tracers (galaxies or halos). We particularly focus on two kinds of patterns: i) clustered-type ones in which the datapoints are separated in the input space into multiple groups. We particularly show that the clustering procedure performed with a Gaussian Mixture Model can be formulated in terms of statistical physics that enables information about the dataset itself; and ii) spatially continuous datasets assumed as standing on an underlying one-dimensional structure that we aim to learn, akin to the filamentary structure. To this end, we resort to a regularization of the mixture model in which a spatial graph is used as a prior to approximate the underlying pattern. Finally, we show that the environments can be used to improve the constraints on cosmological parameters of the LCDM model based on the Quijote suite of simulations. In particular, by breaking some key degeneracies, we report up to an order of magnitude tighter constraints on parameters like the summed neutrino mass and the matter density over the real-space power spectrum.