Neural network

Imagine you are trying to teach a child to recognise a cat without ever writing down a rule. You show her a thousand photos: "cat," "not a cat," "cat." She doesn't memorise a checklist — she absorbs something vaguer and more powerful, a feeling for what cat-ness looks like. A neural network does something structurally similar, except instead of a child, it is a chain of arithmetic operations, and instead of feelings, it is numbers.

That system is called a neural network — a collection of mathematical functions, arranged in layers, that learns to map inputs to outputs by adjusting millions of internal numbers called weights.

How it works

Each layer of a neural network is a set of neurons (also called nodes or units). A neuron does one thing: it takes several numbers in, multiplies each by a weight, adds them up, and passes the result through a simple squishing function. That is it. The power comes from stacking thousands of such neurons across dozens of layers.

During training, the network sees labelled examples. It makes a guess, compares it to the correct answer, measures how wrong it was, and nudges every weight slightly to make the next guess a little better. This process — called backpropagation — repeats millions of times until the network is reliably right.

A real example

The face unlock on a modern smartphone runs a neural network. It compares the infrared image of your face against a compressed mathematical signature it learned when you set up the device. The network was trained on millions of faces so it generalises — it still recognises you with a new haircut, in dim light, or wearing glasses.

Why it matters for robotics

A robot's sensors — cameras, microphones, lidar — produce raw streams of numbers that mean nothing on their own. Neural networks are the layer that turns those numbers into understanding: "this blob of pixels is a person," "this vibration pattern means the joint is slipping," "this voice command means pick up the red block." Without them, a robot is blind and deaf to meaning.

The deeper a neural network gets — the more layers it has — the more surprising and powerful the things it can learn, which is the story of deep learning.