Abstract
One of the approaches used by researchers trying to develop a computer system capable of understanding natural language is that of training a Neural Network for the task. Because of the large number of parameters that can be controlled in a neural network (such as topology, training data, transfer function, learning algorithm, and others) it is not surprising to find that different researchers have used networks with differing configurations, several of which have achieved success with natural language tasks. Despite these successes, little is know regarding how to choose a set of initial configuration parameters for a particular task. In this proposal I will review some of the past efforts by researchers attacking the problem of natural language processing with neural networks. I will then explain how I intend to use genetic algorithms to automatically look for an initial configuration that might achieve a better performance for a particular language task. This method will not only transfer to searching for configurations for different language tasks, but will also generate intermediate information that might result useful in understanding why a particular set of parameters is useful.

Abstract

One of the approaches used by researchers trying to develop a computer system capable of understanding natural language is that of training a Neural Network for the task. Because of the large number of parameters that can be controlled in a neural network (such as topology, training data, transfer function, learning algorithm, and others) it is not surprising to find that different researchers have used networks with differing configurations, several of which have achieved success with natural language tasks. Despite these successes, little is know regarding how to choose a set of initial configuration parameters for a particular task.

In this proposal I will review some of the past efforts by researchers attacking the problem of natural language processing with neural networks. I will then explain how I intend to use genetic algorithms to automatically look for an initial configuration that might achieve a better performance for a particular language task. This method will not only transfer to searching for configurations for different language tasks, but will also generate intermediate information that might result useful in understanding why a particular set of parameters is useful.

Back to the Table of Content

To the next topic