Building a feedforward neural network from scratch

Let's get into a time machine. In nanoseconds, it takes us back to 1969. We have today's knowledge but nothing to prove it. Minsky and Papert have just published their book, Perceptrons. They've proven that a perceptron cannot implement the exclusive OR function XOR.

We are puzzled. We know that deep learning will be a great success in the 21st century. We want to try to change the course of history. Thanks to our time machine, we land in a small apartment. It's comfortable, with a vinyl record playing the music we like! There is a mahogany desk with a pad, a pencil, sharpener, and eraser waiting for us. We sit. A warm cup of coffee appears in a big mug. We're ready to solve the XOR problem from scratch. We have to find a way to classify those dots with a neural network.