deep learning
Deep learning teaches a computer to recognize things the way a child slowly learns to read — not by being handed a rulebook, but by stacking up understanding layer by layer. The "deep" simply means many layers: a neural network with not one or two, but dozens or hundreds of them, each passing what it figured out to the next. Show it a photo, and the first layer might notice only edges and blobs of color; the next assembles those into eyes and wheels; a higher one finally says "cat" or "car."
What makes this special is that nobody tells the network what an "edge" or an "eye" is. It discovers those building blocks on its own, simply by being shown millions of examples and nudged toward the right answers — a process called learning features in a hierarchy. That is the real break from older methods, where humans had to hand-craft the clues a program looked for.
A common mix-up: deep learning is not a separate kind of artificial intelligence sitting beside the rest. It is one powerful technique within machine learning, which in turn lives inside the broad idea of AI. It is not magic, and not a thinking mind — it is pattern-finding at enormous scale, only as good as the examples it is fed.
pixels → edges → shapes → eyes & ears → "cat" — each layer builds on the one below.
How a deep network builds understanding from the ground up, one layer at a time.
The field had simmered for decades, but its modern boom ignited in 2012, when a deep network called AlexNet crushed the field at the ImageNet image-recognition contest — winning by a margin so wide it convinced the world that deep, many-layered networks, trained on graphics chips (GPUs), were the way forward.