Introduction
How powerful is a neural network, really? And if one hidden layer can already do a lot, why do modern networks have dozens or hundreds of layers?
Imagine building shapes with LEGO. With enough tiny bricks laid side by side in a single flat layer, you can approximate any outline — but you'll need an enormous number of bricks. If you can stack them and build reusable parts (wheels, windows, doors), you can build complex things like a whole town with far fewer bricks. Depth gives neural networks the same advantage: reusable, layered building blocks.