In this part of the book, we’ll use what we learned in Part I to build a more sophisticated learning machine: a neural network. Neural networks are leaps and bounds more powerful than perceptrons, but they come with their own set of challenges. We’ll spend a few chapters building our first network, and a few more detailing and overcoming those challenges.
Here’s a mild spoiler: even the first version of our neural network will classify MNIST digits better than the perceptron. But we won’t stop there! As you’ll see soon, we’re going to aim for a target that’s way more ambitious: 99% accuracy in character recognition.
Fasten your seatbelts… this will be quite the trip.