Hebbian learning

Hebbian learning rests on a postulate that the psychologist Donald Hebb set out in his book The Organization of Behavior (1949): when one neuron repeatedly takes part in firing another, the connection between them is strengthened. The catchy summary "cells that fire together, wire together" was coined later by others, but it captures the idea.

The rule was neurophysiological to begin with, but became a cornerstone of artificial neural networks: a simple, local way to adjust weights with no answer key — that is, unsupervised learning. It builds on the McCulloch–Pitts neuron of 1943 and predates the gradient-based methods such as backpropagation that came to dominate. Pure Hebbian learning is unstable on its own — the weights grow without bound — a flaw that later variants had to engineer away.