Created using Visual Studio 2010 in conjuction with IntelOpenCL SDK

This project is currently a basic, fairly optimized OpenCl implementation
of a fully connected feed forward network. In addition, this project has
implemented a single layer convolutional neural network.

Includes ability to store and load neural nets from files

MNIST Data set (Error rates based on the 10k image testing set):
    One fully connected neural net of size 150-200-100-10 achieved an error rate of 3.23%
    Another one of size 800-500-250-10 has achieved an error rate of 2.3%
    Another one of size 1500-1000-500-10 has achieved an error rate of 1.48% 
        after training using simple affine distortions

Added a demo of the neural networks under ./OpenCLNeuralNet/demo/
    -demo1.py Single digit classification
    -demo2.py Multiple digit classification
    The neural network used by the demos is the file "NN-SMALL.net" in
    the same directory as the demo files.

    Required dependencies: Python Imaging Library (PIL) and Numpy