AI glossary · Basic · Language models
Perceptron
A perceptron is the simplest artificial neuron model: it multiplies each of several inputs by a weight, adds them up, and outputs 1 if the sum exceeds a threshold and 0 otherwise. It is the basic unit from which artificial neural networks started.
About this term
- Level
- Basic · Technical terms you often meet in AI news
- Category
- Models & generation
- Fields
- Language models
- Added
- 2026-10-08
How it works
It multiplies each input value by a weight, adds them all together, and then adds an adjustment value called the bias. If the total is greater than 0, it outputs 1; otherwise, it outputs 0. During training, every time it sees an example it got wrong, it nudges the weights toward the correct answer. In this way it finds on its own a boundary line that splits the data into two groups.
Why it matters
The perceptron showed, in a very simple form, the idea that "a machine learns by looking at examples and adjusting its weights." Even a single neuron in today's neural networks has the same basic structure as a perceptron (weighted sum → activation function). That is why it is the first concept covered when people start learning about neural networks.
Things to know
A single perceptron can only solve problems that can be separated by one straight line. A well-known example it cannot solve is XOR, which is "true when exactly one of the two is true." This limitation was overcome by the multilayer perceptron (MLP), which stacks perceptrons in multiple layers, and by backpropagation, which trains it. Today it is usual to use smooth activation functions instead of a hard 0/1 cutoff.
Example
If you built a very simple spam filter, you could imagine a perceptron that takes "whether certain words appear," "number of links" and "whether the sender is in your contacts" as inputs and classifies a message as spam if the weighted sum passes a threshold. Real services use much more complex neural networks, but each neuron inside them computes in this way.