Minimal neural network demo for Octave (Matlab)
Based on equations from the first chapters of M. Nielsen's book
Trains a NN for digit recognition on a set of 5000 20x20 images from the NIST dataset.
- 400 nodes input layer (20x20 pixel)
- 30 hidden nodes
- 10 output nodes (digits 0..9)
Recognition of the training dataset converges to 99.0 % after a minute or two.