Draw number then press space, to identify number. No need for turbowarp (luckily), if you do experience lag remember its 4 layers of neural networks... Uses two convolutional layers of size 3*3, this takes input data and reduces its size and naturalises the inputs, then these are passed through two linear layers of size 120 then 60, with 10 outputs (1 for each number)
This program runs at Olog(n) so is extremely fast. As the number of weights had to be backpropagated before hand the overall data storage is nx^3 so increases exponentially. Trained on MNIST data set which was restructured using java script.
