Picture a digital gatekeeper with just one decision to make: "yes" or "no." It looks at several clues, assigns each a different importance, sums them up, and decides. That is precisely the perceptron: the oldest and most elementary artificial neuron, invented by Frank Rosenblatt in 1958. It is the founding building block of all modern deep learning.
How it works
The perceptron takes inputs $x_1, x_2, \dots, x_n$, assigns each a weight $w_i$ (its importance), then adds a bias $b$. It computes a weighted sum and passes it through an activation function (historically a step: above threshold 1, otherwise 0):
$$ y = f!\left( \sum_{i=1}^{n} w_i x_i + b \right) $$
Learning means adjusting the weights whenever the perceptron is wrong: it is corrected example after example until it classifies correctly.
Strength and limitation
The perceptron can only separate linearly separable data: it draws a single straight line (a hyperplane) between two classes.
| Task | Single perceptron |
|---|---|
| AND / OR function | Solves |
| XOR function | Fails |
Its inability to solve XOR (shown by Minsky and Papert in 1969) stalled research, until it was realized that stacking perceptrons into layers — the multilayer perceptron (MLP) — overcomes this limit.
A single neuron draws a line; thousands of them, arranged in layers, draw today's artificial intelligence.