Concepts

Perceptron

3 min readbeginnerUpdated 28 Sept 2026
1 · In one line

A perceptron is a linear classifier.

1 · What it is

A perceptron is a linear classifier. Different input features can carry different learned weights. A score above zero selects the positive class.

The standard perceptron updates its model only on mistakes. Its decision score is proportional to signed distance from the separating hyperplane.

Rosenblatt presented the perceptron as a probabilistic model for information storage and organization in the brain. The learned decision boundary is linear. By default, scikit-learn’s perceptron is not regularized.

2 · Why it exists

A binary decision needs one repeatable rule for combining several measured features.

Unequal evidenceDifferent input features can carry different learned weights.
Clear boundaryA score above zero selects the positive class.
Online correctionThe standard perceptron updates its model only on mistakes.
3 · How it works

Follow two features into one binary decision.

A score above zero selects the positive class.
  1. 1 · receiveRead an input sample with the configured number of features.
  2. 2 · weightCompute the dot product of the input and kernel.
  3. 3 · sumApply an affine linear transformation.
  4. 4 · decidePredict the positive class when the score is above zero.
  5. 5 · updateDuring training, update the model when it makes a mistake.

A score above zero selects the positive class.

4 · Where it's used
WhoWhat they askWhat it works with
Quality-control team“Does this measurement pattern indicate a pass or fail?”Numeric sensor features
Email filter“Does this message match the positive class?”Numeric message features
Teaching lab“How does an online linear classifier change after an error?”Labelled feature vectors
5 · What it solves, and what it doesn't
solves
  • A perceptron fits a linear model with stochastic gradient descent.
  • Its decision score is proportional to signed distance from the separating hyperplane.
  • The standard perceptron updates its model only on mistakes.
doesn't solve
  • The learned decision boundary is linear.
  • By default, scikit-learn's perceptron is not regularized.
  • Multiclass use trains one binary problem for each class.
6 · Go deeper

Sources used

This explainer is written in original language. The links below support its factual claims.

  1. docsPerceptron, scikit-learn · read 28 Sept 2026
  2. docsLinear Models, scikit-learn · read 28 Sept 2026
  3. docsDense layer, Keras · read 28 Sept 2026
  4. docsLinear, PyTorch · read 28 Sept 2026
  5. paperThe perceptron: A probabilistic model for information storage and organization in the brain, Frank Rosenblatt, Psychological Review · read 28 Sept 2026