This is a demonstration of how a neural network can be used to recognise handwritten digits. Draw a digit in the box below and see how accurately the neutral net is able to recognise it.
Start drawing to see predictions...
Implementation notes: A three layer neural network is used with 784 input neurons, representing a 28x28 pixel image, fully connected to a hidden layer of 100 neurons. The output layer contains 10 fully connected neurons corresponding to each digit from 0 to 9. A softmax activation function is used in the output layer allowing each output to be interpreted as a probability. The model was trained using stochastic gradient descent with the MNIST dataset, available at http://yann.lecun.com/exdb/mnist/. As recommend for this dataset, each image is pre-processed before being passed through the net. This involves first scaling the image to fit in a 20x20 pixel bounding box, then centering the image on its center of mass, and finally normalising each pixel to have a value between 0 and 1. The net was trained in Python and weights and biases were exported to run in Javascript.