Skip to Main content Skip to Navigation
Journal articles

Apprentissage et sélection de réseaux bayésiens dynamiques pour les processus online non stationnaires

Abstract : Dynamic Bayesian Networks (DBNs) provide a principled scheme for modeling and learning conditional dependencies from complex multivariate time-series data. However, in most cases, the underlying generative Markov model is assumed to be homogeneous, mea- ning that neither its topology nor its parameters evolve over time. Therefore, learning a DBN to model a non-stationary process under this assumption will amount to poor predictions capa- bilities. Thus we build a framework to identify, in a streamed manner, transition times between underlying models and a framework to learn them in real time, without assumptions about their evolution. We propose a model for the dynamic of the transitions between modes stemming from Hidden semi-Markov Models (HsMMs) and Graphical Duration Models (GDMs). We show the method performances on simulated datasets.
Complete list of metadata

https://hal-cnam.archives-ouvertes.fr/hal-03228681
Contributor : Research Team Sécurité Défense <>
Submitted on : Tuesday, May 18, 2021 - 4:03:00 PM
Last modification on : Sunday, June 6, 2021 - 3:29:49 AM

File

46 Baumard Hourbracq Revue in...
Publisher files allowed on an open archive

Identifiers

  • HAL Id : hal-03228681, version 1

Citation

Matthieu Hourbracq, Pierre-Henri Wuillemin, Christophe Gonzales, Philippe Baumard. Apprentissage et sélection de réseaux bayésiens dynamiques pour les processus online non stationnaires. Revue des Sciences et Technologies de l'Information - Série RIA : Revue d'Intelligence Artificielle, Lavoisier, 2018, Réseaux bayésiens et modèles probabilistes, 32 (1), pp. 75-109. ⟨hal-03228681⟩

Share

Metrics

Record views

395

Files downloads

43