perceptron
/ per-SEP-tron /
A perceptron is the simplest artificial neuron: a tiny decision-maker that looks at several inputs, gives each one a weight saying how much it matters, adds them all up, and then flips a switch — yes or no — depending on whether the total clears a threshold. Picture a doorman deciding whether to let you in: height counts a little, the guest list counts a lot, and once the points add up past the cutoff, the answer is simply in or out.
What made it electric in 1958 was that it could learn. Show it examples labelled "yes" and "no", and every time it guesses wrong it nudges its weights a notch in the right direction. Do this enough times and it settles on a set of weights that draws a clean line between the two groups — handwritten apples on one side, oranges on the other — with nobody having programmed the rule by hand.
But there's a catch that became famous. A single perceptron can only separate things a straight line can split. Ask it to learn XOR — "yes when exactly one of two switches is on" — and it fails, because no single straight line can carve out that pattern. That limit chilled the field for years, until people stacked many neurons in layers and found a way to train them — which is exactly what modern neural networks do.
Weigh the inputs, add them up, and fire a 1 only if the sum clears the threshold.
Coined by psychologist Frank Rosenblatt, who built the Mark I Perceptron in 1958 — a room-sized machine wired to learn from examples. In 1969 Marvin Minsky and Seymour Papert proved a single-layer perceptron could not learn XOR; the result is widely credited with stalling neural-network research for over a decade.