Modelos ocultos de markov en espacios de disimilaridad: alternativas para la selección de prototipos
Abstract
El criterio convencional de clasificación en sistemas que involucran modelos ocultos de Markov emplea la regla de máxima verosimilitud para escoger la clase correcta. Existe evidencia que muestra que la clasificación basada en disimilaridades entre modelos ocultos de Markov aumenta el desempeño del sistema. En este nuevo espacio de disimilaridades, las reglas de decisión pueden construirse usando todo el conjunto de entrenamiento o un conjunto reducido de prototipos adecuadamente seleccionados, que permiten minimizar el número de disimilaridades que deben medirse. En este artículo se comparan diferentes procedimientos para la selección de prototipos en el espacio de disimilaridades entre modelos ocultos de Markov.Downloads
Downloads
-
Vistas(Views): 228
- PDF (Español (España)) Descargas(Downloads): 295
Published
How to Cite
Issue
Section
License
Copyrights
The journal is free open access. The papers are published under the Creative Commons Attribution / Attribution-NonCommercial-NoDerivatives 4.0 International - CC BY-NC-ND 4.0 license. For this reason, the author or authors of a manuscript accepted for publication will yield all the economic rights to the Universidad Tecnológica of Pereira free of charge, taking into account the following:
In the event that the submitted manuscript is accepted for publication, the authors must grant permission to the journal, in unlimited time, to reproduce, to edit, distribute, exhibit and publish anywhere, either by means printed, electronic, databases, repositories, optical discs, Internet or any other required medium. In all cases, the journal preserves the obligation to respect, the moral rights of the authors, contained in article 30 of Law 23 of 1982 of the Government Colombian.
The transferors using ASSIGNMENT OF PATRIMONIAL RIGHTS letter declare that all the material that is part of the article is entirely free of copyright. Therefore, the authors are responsible for any litigation or related claim to intellectual property rights. They exonerate of all responsibility to the Universidad Tecnológica of Pereira (publishing entity) and the Scientia et Technica journal. Likewise, the authors accept that the work presented will be distributed in free open access, safeguarding copyright under the Creative Commons Attribution / Recognition-NonCommercial-NoDerivatives 4.0 International - https://creativecommons.org/licenses/by-nc-nd/4.0/deed.es license.