Handwritten Digit Recognition with Neural Network

The network get a drawing digit and convert it to a B/W smaller image (28x28)
Each pixel is fed into an input layer neuron which propagates through the network to get a digit classification on the right.
The hidden layer have only 25 neurons with a sigmoid activation function.
The output layer uses softmax to provide a probability classification function.
"clear" the drawing area, draw digit and click on "class" for classification.
To train the network enter a number in the textbox and click "train". To accelarate the learning the traning will generate 9 more distorted images to train with.
The page is loaded with some trained network. To start from scratch, click on "ClearNet".


Probabilities: