Skip to Main content Skip to Navigation
Theses

Contributions aux Modèles de Markov Cachés : métaheuristiques d'apprentissage, nouveaux modèles et visualisation de dissimilarité

Abstract : In this PhD thesis, we present many contributions aimed at the improvement on the utilization of hidden Markov Models (HMMs) in artificial intelligence systems. We considered three main objectives :
improving HMM training, experimenting a new HMM and visualizing interaction among HMMs. In the first part, we propose, evaluate and compare many new adaptations of classic biomimetic meta-
heuristics (genetic algorithms, API artificial ants algorithm, particle swarm optimization) applied to the problem of HMMs training. In the second part, we propose a new kind of HMM which we named symbols substitution hidden Markov models (SSHMMs). A SSHMM allows an expert to incorporate a priori knowledge in the training and the recognition tasks. First experiments with such models show that SSHMMs are of great interests at least for images learning and recognition. In the third part,
we propose a new visualization technique to tackle dissimilarity. This technique, which we named the pseudo-euclidean scatterplots matrix (PESM), allows a better understanding of interaction between
HMMs. This PESM is built from techniques which we named indefinite kernel principal component analysis (IKPCA). Finally, our research concludes with the description of the HMMTK software library
developed along this work. The library integrates parallelization mechanisms and algorithms developed during the thesis.
Document type :
Theses
Complete list of metadatas

Cited literature [7 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00168392
Contributor : Sébastien Aupetit <>
Submitted on : Monday, August 27, 2007 - 7:26:30 PM
Last modification on : Tuesday, November 19, 2019 - 4:46:42 PM
Long-term archiving on: : Monday, September 24, 2012 - 11:46:07 AM

Identifiers

  • HAL Id : tel-00168392, version 1

Collections

Citation

Sébastien Aupetit. Contributions aux Modèles de Markov Cachés : métaheuristiques d'apprentissage, nouveaux modèles et visualisation de dissimilarité. Interface homme-machine [cs.HC]. Université François Rabelais - Tours, 2005. Français. ⟨tel-00168392⟩

Share

Metrics

Record views

517

Files downloads

7080