Use this link to cite:
http://hdl.handle.net/2183/31917 Aprendizaxe por currículo no adestramento de redes profundas
Loading...
Identifiers
Publication date
Authors
Mato González, Daniel
Other responsabilities
Universidade da Coruña. Facultade de Informática
Journal Title
Bibliographic citation
Type of academic work
Academic degree
Abstract
[Resumen]: La idea de ordenar tareas o datos dependiendo de su dificultad, creando un currículo, se lleva
usando muchos años en ámbitos como el de la enseñanza y, más recientemente, el aprendizaje
máquina. En este trabajo aplicaremos el uso de un currículo en aprendizaje profundo,
comparando los resultados obtenidos con varios tipos de aprendizaje por currículo y un caso
base con una red neuronal profunda entrenada sin utilizar ningún currículo. Para comprobar
su eficacia, usaremos las redes neuronales resultantes para clasificar fragmentos de polisomnogramas
en cinco clases, representando las cinco etapas del sueño.
[Abstract]: The idea of sorting tasks or data by their difficulty, creating a curriculum, has been used for a long time in fields like teaching and, more recently, machine learning. In this study we apply the use of a curriculum to deep learning, comparing the results obtained through different types of curriculum learning and a base case with a deep neural network trained without using a curriculum. In order to test their effectiveness, we will use the resulting neural networks to classify polysomnogram fragments in five classes, representing the five sleep stages.
[Abstract]: The idea of sorting tasks or data by their difficulty, creating a curriculum, has been used for a long time in fields like teaching and, more recently, machine learning. In this study we apply the use of a curriculum to deep learning, comparing the results obtained through different types of curriculum learning and a base case with a deep neural network trained without using a curriculum. In order to test their effectiveness, we will use the resulting neural networks to classify polysomnogram fragments in five classes, representing the five sleep stages.
Description
Editor version
Rights
Todos os dereitos reservados







