The Feynman Companion

Part 3

Inside the Neural Network

From a Single Neuron to a Factory Inspector

The black box, opened.

"What I cannot create, I do not understand."

— Richard Feynman

Prologue: What We’re Opening

In Part 1, we followed photons through cameras and USB stacks. In Part 2, we followed pixels through preprocessing, tiling, and production loops. Both parts treated the neural network as a black box: something goes in, a decision comes out. Now we open it.

We’re going to build a neural network from scratch in your head—starting with a single neuron, then stacking them into layers, then arranging layers into the specific architectures that inspect factory parts. By the end, you’ll understand not just what these models do but WHY they’re shaped the way they are. Every design choice is an answer to a specific question about how to see.

Chapter 1: The Neuron