Keeping Neural Networks Simple

Research on minimizing description length of weights for better generalization.