Can we think of neural networks as layers of decisions with regression and classification at each layer?