Some ideas on improving generalization

a nice sum up of ideas how to improve generalization of a neural network

Any other ideas?