Handwritten Digit Recognizer

1
2
3
4
5
6
7
8
9
0


Draw a digit from 0 to 9 in the gray box above, and press clear before drawing another digit. This detection is done through a 400x25x10 artificial neural network. The ANN was trained with a subset of the MNIST handwritten digit data, via backpropogation in Matlab. The drawing area is 20x20 pixels, leading into 400 input neurons. The hidden layer is 25 neurons and the output is 10 neurons, one for each possible digit. The activation levels of each output neuron is displayed above. The neural network is executed in JavaScript, and uses two csv files as weights for the edges of the network. My next step is to reverse the neural network, and allow the network to "dream" of digits. We would input a digit as text, and the network would iteratively draw that digit.