The code for a single neuron is fairly simple. When we combine neurons, however, the input is passed through multiple neurons in a hidden layer, which can be very large. The output of the hidden layer is itself either passed to more layers or an output layer of variable size. This can involve absolutely huge computations which are hard to understand and code efficiently.

To understand these computations and work with neural network libraries, you must refresh your linear algebra and be able to think of networks in terms of matrix calculations. We'll warm you up with this gentle exercise.