This is the first chapter of 3Blue1Brown’s deep learning series, created and animated by Grant Sanderson on his own channel. Using the concrete task of recognizing handwritten digits, it builds up, from first principles, what a neural network is and how information flows through it.
Sanderson introduces neurons, layers, weights, and biases visually, showing how a network transforms the pixels of an image into a prediction about which digit it represents. Rather than relying on analogies to the brain, he keeps the explanation grounded in the actual math, but renders that math so clearly that a viewer with no background can follow it. The video sets up later chapters on how such networks learn through gradient descent and backpropagation.
This is original, firsthand teaching, not a recap of someone else’s material, and it is frequently cited as the single best place to start for anyone who wants real intuition about neural networks. For a reader new to the field, it pairs naturally with the more advanced builds and lectures elsewhere in this collection.