Hi - in this slide, Professor Ng refers to a way to choose a neural network architecture (slide below). What determined the options for the initial layers, neurons etc.? why cant they be 3 layers, 1 neuron each, or 40 layers 50 neurons each? Thanks for the explanation!!
The size of the input and output layers is fixed by the dataset (the number of features) and labels (the number of outputs).
The size of the hidden layer (and the number of hidden layers) is a free variable that you select by experimentation.
Thank you!