Estimation bayésienne asymptotique de la structure d'un graphe initialisée par Graphical lasso
Résumé
When studying process with multivariate time series, a point of interest is the knowledge about conditional dependency. For Gaussian times series, Gaussian graphical models are commonly used to represent such dependencies. In this paper, we present an approach that estimates the graph structure representing the conditional dependencies of a process. This approach uses Graphical lasso to fasten a Bayesian approach. The obtened solutions for datasets simulated from known graph structures are closed, according to the Hamming distance, to the expected solution. Moreover, our approach has a lower computing cost than a Bayesian exhaustive research.