Programming Machine Learning

Assembling a Neural Network from Perceptrons

Let’s see how to build a neural network, starting with the perceptron that we already have. As a reminder, here is that perceptron again—a weighted sum of the inputs, followed by a sigmoid:

In Part I of this book, we didn’t just use the perceptron as is—we also combined perceptrons in two different ways. First, we trained the perceptron with many MNIST images at once; and second, we used ten perceptrons to classify the ten possible digits. In ​Assembling Perceptrons​, we compared those two operations to “stacking” and “parallelizing” perceptrons, respectively, as shown in the picture.

To be clear, ...