A Statistical Approach to Neural Networks for Pattern Recognition

by Dunne, Robert A.

• ISBN13: 9780471741084
• ISBN10: 0471741086
• Edition: 1st
• Format: Hardcover
• Copyright: 2007-07-16
• Publisher: Wiley-Interscience

Summary

This book presents a statistical treatment of the Multilayer Perceptron (MLP), which is the most widely used of the neural network models, in a language that is familiar to practicing statisticians. Questions arise when statisticians are first confronted with such a model, and this book's aim is to provide thorough answers. The following are a few questions that are considered in this book and are explored: how robust is the model to outliers, could the model be made more robust, which points will have a high leverage, what are good starting values for the fitting algorithm, etc. Discussions include the use of MLP models with spatial data as well as the influence and sensitivity curves of the MLP. The question of why the MLP is a (fairly) robust model is answered and modifications to make it very robust are considered.

Author Biography

Robert A. Dunne, PhD, is Research Scientist in the Mathematical and Information Sciences Division of the Commonwealth Scientific and Industrial Research Organization (CSIRO) in North Ryde, Australia. Dr. Dunne received his PhD from Murdoch University, and his research interests include remote sensing and bioinformatics.

Table of Contents

Preface

Acknowledgments

Introduction

The Multi-Layer Perception Model

Linear Discriminant Analysis

Activation and Penalty Functions

Model Fitting and Evaluation

The Task-Based MLP

Incorporating Spatial Information into an MLP Classifier

Influence Curves for the Multi-Layer Perceptron Classifier

The Sensitivity Curves of the MLP Classifier

A Robust Fitting Procedure for MLP Models

Smoothed Weights

Translation Invariance

Fixed-slope Training

Appendix A

Appendix B

Topic Index

Table of Contents provided by Publisher. All Rights Reserved.