Optimal Input Gain: All You Need to Supercharge a Feed-Forward Neural Network