Christophe Ambroise

Contrast-Free ICA and Causal Inference via Wasserstein Distances to the Gaussian

Félix Laplante, Christophe Ambroise, Pierre Humbert

We study the squared $2$-Wasserstein distance to the standard Gaussian as a non-Gaussianity criterion and use it for linear Independent Component Analysis (ICA) and causal discovery in Linear Non-Gaussian Acyclic Models (LiNGAM). Unlike commonly used parametric contrasts and appr…Read more