Título de la tesis:
Aprendizaje multitarea en problemas con un número reducido de datos
Autor/es:
Bueno Crespo, Andrés - Sancho Gómez, José Luis
Tipo de documento:
Ponencia
Universidad:
Universidad Politécnica de Cartagena
Departamento:
Idioma:
Castellano
Palabras clave:
Aprendizaje multitarea, Redes Neuronales Artificiales (RNA), Tareas artificiales, Edición de datos
Fecha de la defensa:
Sep-2005
Notas:
MultiTask Learning (MTL) is a procedure to
train a neural network to learn several related tasks
simultaneously considering one of them as main task and
the others as secondary tasks. In this paper, we have
tested a method to obtain artificially tasks which are
related with the main one, because in many real cases,
knowledge about problem to be solved is uncertain.
We use sample selection techniques to generate related
tasks with the main one, in particular, samples close the
classification boundary. Moreover, a new procedure to
train MultiLayer Perceptrons with generated tasks is
d...
Valoración:
Clasificación por
