Alessandro Antonucci, Rocco De Rosa, Alessandro Giusti, Fabio Cuzzolin

---

Classification of Temporal Data by Imprecise Dynamic Models

Abstract

A new procedure to classify temporal data with im- precise hidden Markov models is proposed. A differ- ent model is learned from each sequence by coupling the imprecise Dirichlet model with the EM algorithm. As a descriptor of the model associated to a sequence, we consider the expected value of the manifest vari- able in the limit of stationarity of the Markov chain. For imprecise models, only the bounds of this descrip- tor can be evaluated. In practice, the sequence, which can be regarded as a trajectory in the features space, is summarized by this method as a hyperbox in the same space. These static but interval-valued data are classified by a credal extension of the k-NN algorithm. Experiments on human action recognition data show that the method achieves the required robustness and outperforms other imprecise methods.

Keywords

Credal networks, Imprecise Hidden Markov Models , Credal classification

---

Download area

The paper is available in the following formats:

• pdf file

E-mail addresses

>

Alessandro Antonucci		alessandro@idsia.ch
Rocco De Rosa		roccomilano@tiscali.it
Alessandro Giusti		alessandrog@idsia.ch
Fabio Cuzzolin		fabio.cuzzolin@brookes.ac.uk