Video narration
How does a neural network learn without being a brain or a black box? Think of trying a recipe. You choose ingredients, bake, and taste. A network takes an example as its ingredients: input values pass through connected calculations, producing a guess. In supervised learning, that guess is checked against a known answer, like tasting your batch against the result you wanted. The difference is the error. Here's the turn: a wrong guess isn't wasted. It guides small changes to the network's weights, the strengths of its connections, much as feedback tells you what to adjust in the recipe. The comparison is only an analogy; calculations make the network's changes. It repeats this training across examples. Over time, guesses improve, and the network can make a guess on a new example. So how does it learn? It guesses, checks, adjusts, and repeats-like refining a recipe.