This is a simple neural network based ocr experiment.
Training data in ./letter.data
is taken from ai.stanford@btaskar
The process.js file turns the raw training data into a js-friendly format
- Uses 128 input neurons á 8x16 pixels
- Uses 125 hidden layers (really not that fine grained)
- The 7 output layers return a binary encoded letter as the activation result