Publication:
The use of backpropagating neural networks in coordination with logistic spline coefficients to obtain the change-point

Thumbnail Image
Authors
Maldonado-Colberg, Elisa M.
Embargoed Until
Advisor
McGee, Daniel L.
College
College of Arts and Sciences - Sciences
Department
Department of Mathematics
Degree Level
M.S.
Publisher
Date
2006
Abstract
This work presents an algorithm that uses backpropagating neural networks and their ability to find logistic regression coefficients as a useful tool that may outperform other conventional methods for finding the change-point of a data set. To seek this objective, we first demonstrate the capacity of a one-layered neural network to find logistic regression coefficients. Then, we add another layer to the network in order to determine if two-layer discriminants would allow us to approximate the change-point. We then develop an algorithm based on hidden layer coefficients of neural networks that allows us to assess change-points based on when a neural network switches from one “neuron” to another.

Este trabajo presenta un algoritmo que utiliza redes neurales artificiales de retropropagación y su habilidad para encontrar coeficientes de regresión logística como herramienta útil que puede superar métodos convencionales para hallar el punto de cambio en un conjunto de datos. Para lograr este objetivo, primero se demuestra la capacidad que tiene una red neural de un nivel para encontrar coeficientes de regresión logística. Entonces, añadimos otro nivel adicional a la red neural para determinar si los dos discriminantes que ésta calculó nos aproximan al punto de cambio. Luego, se creó un algoritmo basado en coeficientes de un nivel escondido de la red que nos ayudará a encontrar los puntos de cambio basándonos en el momento en que la red se mueve de una neurona a otra.
Keywords
Backpropagating neural networks
Cite
Maldonado-Colberg, E. M. (2006). The use of backpropagating neural networks in coordination with logistic spline coefficients to obtain the change-point [Thesis]. Retrieved from https://hdl.handle.net/20.500.11801/1987