Neural Network

ANNemarie is an Artificial Neural Network capable of learning and recognizing characters from 8×8 bitmaps. It contains a matrix visualization component which allows the user to see changes made to the matrix in real time, as well as a graphing component that displays the rate of learning of the neural network. A step mode function is also available for users who wish to see the neural network training example by example. ANNemarie is written in C# using Microsoft Visual Studio 2008.

Each example training set contains 100 examples and is made up of bitmap representations of the digits 0 to 9. However, the user can create his own training sets if desired. Note that the more times a example appears in the training set, the wider the margin of error of the test input, but the longer it will take for the neural network to train successfully.

The learning rate (mu) and error threshold (delta) are both set to 0.2, but may be varied if necessary.

To see the neural network in action, load one of the example training sets and press Start. Once the neural network has trained successfully, click on the Test tab to see whether the neural network can recognize variations of the digits it has learned.

Download ANNemarie.

Download the example training sets.

Download the documentation.

Known Errors

The neural network does not yet support:

  • multiple layers
  • non-numeric training sets
  • bitmaps larger than 8×8
  • graphing results for epochs over 1000

Leave a Reply