Perceptron
An early artificial neuron (1958) that mimicked biological neurons by combining weighted inputs and firing if they exceeded a threshold.
Think of it like
A simple light switch that turns on if the total incoming signal is strong enough.
Example
A perceptron could learn to classify handwritten digits or simple patterns by adjusting its weights: if it misclassified, it would tweak the weights slightly and try again.
How it actually works
Frank Rosenblatt's perceptron was the first learning machine that provably converged: given enough examples, it could learn any linearly separable classification task. It sparked huge optimism — scientists predicted AGI was near — but Minsky and Papert showed in 1969 that single perceptrons couldn't solve even simple problems like XOR. This led to the first "AI winter." Perceptrons were resurrected as building blocks of multilayer neural networks, which could solve nonlinear problems. Today's deep networks are perceptrons stacked in layers; the perceptron remains conceptually central to how neural networks work.
For product teams
An early success that raised unrealistic expectations, then faded until neural networks scaled to practical usefulness.
For engineers
A simple linear classifier: output = 1 if sum(w_i * x_i + b) > 0, else 0. Learns by adjusting weights via the perceptron update rule.
Related
- Connectionism — Foundation of connectionism and neural networks.
- Neural network — A building block of deep learning.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome