Implement maxout layer

[subversive-owl]

See Goodfellow 2013 for paper.

[drahnr]

I had a quick look on it, but it looks like max pooling in one dimension which are refered to as different modesl? I have yet to sit down and dig into the details.

[Amanita-muscaria]

Max pooling is moving a window over the input data and taking the maximum value at each step. A maxout neuron does inputs * weights and then outputs the maximum value of the result. So a max pool in a convolutional network would be equivalent to a maxout layer with all weights set to one and none of the weights are trainable