Neural Network Hypercubes

I have been thinking about the problems associated with spatial modeling and feature detection to shed some light into the ideas in the previous post. There is a good article on a Face Detection and Recognition (FDR) system at http://factaee.elfak.ni.ac.yu/fu2k71/9caleanu.pdf. However,

In the field of pattern recognition, the combination of an ensemble of classifiers has been proposed to achieveimage classification systems with higher performance incomparison with the best performance achievable employing a single classifier. quote from http://www.cipprs.org/vi2002/pdf/s6-2.pdf

The idea of extracting different feature domains from 2D images led to my desire to revisit my box of neural network papers. I am thinking that I can combine these models with the Kalman Filters of the previous post. In the quote above, the authors set the stage for using a hybrid N-Feature Neural Network (HNFNN) structure that uses radial basis function neural networks for three feature domains.

There are several good articles on using neural networks in C#.NET. A three part series on AI: Neural Network for Beginners at http://69.10.233.10/KB/recipes/Backprop_ANN.aspx . Another example, this one at http://www.codeproject.com/KB/recipes/aforge_neuro.aspx that enables time series prediction for a univariate time series. A neural network OCR at http://www.codeproject.com/KB/cs/neural_network_ocr.aspx as well as the work of http://windale.com/mlpx.php . In the course of my research, I am on the path with the source code for a Face Matching Demo at http://franck.fleurey.free.fr/FaceDetection/demo.htm. This has all the essentials for starting from scratch and implementing models at http://franck.fleurey.free.fr/NeuralNetwork/index.htm. The code is constructed with classes for ActivationFunction.cs, Layer.cs, Neuron.cs, LearingAlgorithms.cs, and neuralnetwork.cs. I recommend using this as a basis to implement the ideas in the above papers and for extension to problem areas.

A good place to start for Java is at http://fbim.fhregensburg.de/~saj39122/jfroehl/diplom/e-index.html . Another is Java Object Oriented Neural Engine (JOONE) at http://www.jooneworld.com/docs/sampleEngine.html . As far as applets go for demos, check out http://neuron.eng.wayne.edu/software.html .

Back to thinking in terms of five dimensions with

lends itself naturally to neural network models with multiple hidden layers and an extension to the basic ideas expressed above.