Implement Weight / Activation Regularization

[fabioseel]

Moved over from other project, slight adjustments & improvements.

Pretty straightforward usage, simply modify you training loop / loss calculation as follows:

model = # define your model

########################
reg = WeightRegularization(model, p=2, weight_decay=0.01) # register model to the regularizer
########################

output = model(...)
loss = # calculate your loss

########################
loss += reg.penalty() # Just add the additional penalty to the loss
########################

# Proceed as usual
loss.backward()