Agenda

25 Mar 2026 12:15

Hans Montcho (King Abdullah University of Science and Technology - KAUST)

Sala Partesotti, San Giobbe Economics Campus + online

Hans Montcho (King Abdullah University of Science and Technology - KAUST) - Skewed posterior approximations for latent Gaussian models using orthogonal polynomials and variational Bayes

Abstract:

Recent advances in approximate Bayesian computation have focused on skewed distribution families that extend Gaussian approximations, supported by skewed Bernstein–von Mises results. We propose a fully deterministic Variational Bayes approach to refine any Gaussian approximation obtained via Laplace’s method, Variational Inference, or Expectation Propagation. Inspired by the skew-normal distribution, our method applies element-wise orthogonal polynomial mappings that preserve joint dependence while modifying the marginal distributions. The cornerstone of the method is the analytical tractability of expectations of polynomial transformations of Gaussian variables, yielding an efficient trade-off between expressivity, scalability, and numerical stability.

The seminar can be attended also remotely, connecting to ZOOM.

Link Zoom: bit.ly/insem-2425
ID riunione:  880 2639 9452
Passcode: InSem-2425

Lingua

L'evento si terrà in inglese

Organizzatore

Department of Economics (InSeminars)

Cerca in agenda