What Is a Neural Network

A neural network is a big adjustable function that learns by guessing and correcting. It takes numbers in (say, the pixels of an image), passes them through layers of simple math, and outputs numbers (say, 'how cat-like is this?'). Each connection has a weight, a dial that says how much one signal matters. At first the dials are random, so the guesses are garbage. You show it an example with the right answer, measure how wrong it was (the 'loss'), and nudge every dial a little in the direction that would have reduced the error. That nudging step is called backpropagation, and it's really just calculus finding which dials to turn. Repeat over millions of examples and the dials settle into values that turn raw pixels into good answers. Nobody programs the rules; the network discovers them by minimising mistakes.