We've defined the input, that's the easy part. Now we need to decide on a network architecture. How can we know how many layers, and how many neurons we should include? I'd like to give you a formula. I really would. Unfortunately, it doesn't exist. In fact, some people are trying to build neural networks that can learn optimal architectures for other neural networks. For the rest of us, we will have to either experiment, search for ourselves, or borrow someone else's architecture.