Drawing ConvNets (just for fun)

Hate overfitting? Well, here overfitting is finally 'useful'! Three ConvNets with 8 hidden layers are treating the pixels of an image as a learning problem: they take the (x,y) position on a grid and learn to predict the color at that point using regression to (r,g,b). We are comparing three different activator functions (ReLU, maxout and tanh) and see how fast they overfit (it trains live in your browser!). Which Neural Net is your favorite artist?

Written in nodejs using browserify convnetjs