Neural network
A system of simple computing units called neurons, connected in many layers, that learns from data how strong each connection should be and so becomes able to make complex judgements.
Each neuron merely multiplies its inputs by weights, adds them up and passes the result to the next layer. Stack many layers and the network can do complicated things such as recognising images or generating text. Backpropagation, a way of working backwards from the output to find how each weight should change, was popularised by David Rumelhart and colleagues in 1986 and made it possible to train networks with several layers[1].
In the 2010s, abundant data and computing power let much deeper networks succeed, and this deep learning became the basis of today's AI[2]. Because the weights are adjusted automatically from examples rather than written as rules, even the people who built a network find it hard to say why it gives the answers it does.
Related terms
- NeuronThe smallest unit of computation in a neural network. It weights and sums its inputs and produces an output depending on the result.
- ParameterAn adjustable number that determines how a model or a machine behaves. In AI it means the strengths of the connections between neurons, the weights, which are set by training.
- Large language modelAn AI trained on vast amounts of text that generates language by predicting the next word. The core of ChatGPT, Claude and similar systems.