How can the different units in a neural network layer compute different outputs when they are receiving the same inputs from the previous layers? If they are receiving the s