Categorizing Skylines With Neural Networks Essay
Summary of relevant material
Nils J. Nilsson, Mathematical Foundations of Learning Machines, Morgan Kaufmann, 1990.
Chapters read: 1-3
The first chapter presents an overview of some of the many applications of neural networks for pattern recognition, and presents a basic model for trainable pattern classifiers. Put simply, these pattern classifiers take in some n number of inputs, and combine them according to m discriminators to output one of m different categories. Since the combinations of different possible inputs is astronomical, it is necessary for the network to correctly categorize novel inputs.
The simplest discriminant function is to simply linearly multiply each input into each node by a number referred to as a synaptic weight, and then sum them. Whichever sum is the highest will be chosen as the output category. A variety of different discriminant functions may be used, which are described at length in the second chapter.
There are two general types of training methods that are used. Parametric methods are appropriate when the categories are known to be correlated with certain parameters, and discriminant functions can be estimated prior to training, while nonparametric methods are useful when the correlation is not intuitive or obvious.
Christopher M. Bishop, Neural Networks for Pattern Recognition, Oxford, 1995.
Chapters read: 1
Artificial neural networks have long been utilized for pattern recognition,…