A line in 2D moves to separate red and blue points, one mistake at a time.
From the chapter: Chapter 1: What Is AI?
Glossary: perceptron, linear classifier
Transcript
The perceptron is the simplest learning machine. Rosenblatt described it in 1958.
Here are some red and blue points in two dimensions. The goal is to find a straight line that puts red on one side and blue on the other.
The perceptron starts with a random line. Most of the points are on the wrong side.
For each misclassified point, the rule is simple. Move the line toward the point if it is on the wrong side, and away if it is on the right side. The amount of movement is fixed by a learning rate.
Watch the line walk through the data. Every time it gets a point wrong, it nudges itself a little closer to fixing that point.
After a few passes, the line settles. Every red point is on one side, every blue point on the other. The data are linearly separable, and the perceptron has found the boundary.
This rule is the ancestor of every gradient-based learning algorithm in modern AI.