⁹One could alternatively view the set of weights, directed to a given layer and coming from one particular neuron in the preceding layer, as a branched axon for the output of that particular neuron. Then we would no longer need the equivalent of dendrites, and we could relabel the weights as belonging to neurons in the preceding layer. All this would not make any difference to the network functionality: it merely concerns the way we wish to denote and distinguish for ourselves the different components of a neural network.