What is a neural network in simple terms?

A neural network is a web of tiny decision-makers called neurons, loosely inspired by the brain. Each neuron takes in some signals, weighs how important each one is, adds them up, and passes a result onward. Stacked together in layers, these simple units can learn to recognise complex patterns like faces, speech, or text.

What are weights in a neural network?

Weights are numbers that say how much each input matters to a neuron. A big weight means that clue strongly influences the decision; a small or negative weight means it barely matters or counts against. Learning, in a neural network, is simply the process of adjusting all these weights until the answers come out right.

How does a neural network learn?

It makes a guess, compares it to the correct answer, and measures how wrong it was — that gap is called the loss. Then it nudges every weight a little in the direction that would have reduced the error. Repeat this over many examples and the weights gradually settle into values that produce good answers.

Is a neural network really like a brain?

Only loosely. The word neuron borrows from biology, and the idea of many simple units connecting was brain-inspired. But artificial neurons are just arithmetic — multiply, add, and a threshold — and a real brain is vastly more complex. The analogy is helpful for intuition, not a literal description.

Neural Networks ELI5: Teaching Computers to Recognise Patterns

What a neural network really is

A neural network sounds mysterious, but the core idea is simple: it is a big web of tiny decision-makers called neurons, each doing a little bit of arithmetic. A single neuron takes in some clues, decides how much each clue matters (those importance numbers are called weights), adds everything up, and passes the result along. Connect thousands or millions of these simple units in layers and the whole network can learn to recognise faces, understand speech, or write text. The interactive demo below lets you play with one neuron deciding “cat or not cat.”

How it works: clues, weights, and a threshold

Tick the clues in the demo — whiskers, meow, four legs — and each becomes an input signal. The neuron multiplies each clue by its weight and sums the results into a single score. If the score clears a threshold, the neuron fires “yes, cat”; otherwise “no.” Crank up the weight on “says meow” and that clue starts to dominate the decision; turn a weight negative and the clue actually argues against cat. This tiny mechanism — weighted sum plus a threshold — is the building block repeated everywhere in real networks.

How it learns from mistakes

A fresh neural network is hopeless: its weights start more or less random, so it guesses wrong constantly. Learning is the process of fixing those mistakes. When the neuron gives the wrong answer, we measure the error and nudge each weight a little in the direction that would have made the answer more correct — weights for clues that should have pushed toward “cat” go up, misleading ones go down. In the demo, press Learn from mistake and watch the weights shift toward the right answer. Do this across thousands of examples and the network slowly teaches itself.

Tips for building intuition

A few things make neural networks click. First, a single neuron is dumb — its power comes from stacking many in layers, where early layers spot simple features (edges, sounds) and later ones combine them into complex ideas (faces, words). Second, the network never gets told the rules; it only sees examples and errors, and discovers the rules by adjusting weights. Third, “training” just means repeating the guess-measure-nudge loop an enormous number of times. Play with the demo until you can predict how a weight change will move the score — once that feels obvious, you understand the heartbeat of every AI model, from the smallest classifier to the largest language model.