One of the major difficulties while working with cosmological
observations in non-linear regimes is to efficiently characterize their
non-Gaussian features. This is for example necessary while studying the
late time Large Scale Structure (LSS), or when modeling the CMB
foregrounds. To study those highly non-Gaussian fields, no generic tool
using high-order correlation functions has clearly emerged.
Introducing an alternative method, I will present a new statistical
description of non-Gaussian fields, which have been developed in data
science and modified to be applied to physical processes. This method,
that computes the covariance of phase harmonics of wavelet coefficients,
is especially suited to statistical characterize the relative phase
shifts between different scales. I will show how this method can be used
to infer cosmological parameters from LSS maps, as well as to produce
realistic statistical syntheses of such maps. I will also briefly show
how these tools can be applied to the study of the galactic CMB foregrounds.