A New Multivariate Time Series Co-clustering Non-Parametric Model Applied to Driving-Assistance Systems Validation
Un nouveau modèle de coclustering non-paramétrique appliqué à la validation de systèmes d'aide à la conduite
Résumé
In this paper, we propose a new Bayesian co-clustering approach applied to Multivariate time series. Our methodology of Functional Non-Parametric Latent Block Model (FunNPLBM) simultaneously creates a partition of observation and a partition of temporal variables, using latent multivariate gaussian block distributions. We propose to use a bi-dimensional Dirichlet Process as a prior for the block distributions parameters and for block proportions, which natively provides model selection. This approach is benchmarked and studied on a simulated dataset and applied to an advanced driver-assistance system validation use-case.
Origine : Fichiers produits par l'(les) auteur(s)