0 avis
Calcul de probabilités et estimation dans des modèles de Markov cachés graphiques
Archive ouverte
Edité par CCSD -
National audience. We show how to define a class of hidden Markov models based on graphical models. Efficient and numerically stable algorithms are derived, that allow to compute the smoothed probabilities involved in the EM algorithm for parameter estimation, and are based on intermediate quantities with explicit probabilistic interpretation.