The author introduces a novel nonparametric ADVI framework using spline approximations to approximate posterior distributions, achieving flexibility, parsimony, and interpretability.
Effiziente Posterior-Approximation durch nichtparametrische Spline-Approximation.