← CS7150 class demos

Rosenblatt's perceptron: the update, the cone, and the bound

Perceptron learning in data space and weight space Left: labelled points, the margin corridor, the model's separating plane, and the update parallelogram. Right: weight space with one constraint line per example, the green wedge of weight vectors that classify everything correctly, the blue bound cone, the trajectory so far, and an optional worst-case zigzag riding the cone edge. A slider scrubs backwards and forwards through the run's history.
points
history 0
Amber line: the model's separating plane, perpendicular to w. Blue cone: where w must lie after k mistakes. Green wedge: every w that classifies every point correctly. Pink zigzag: the extremal run, riding the cone edge. Each step picks adversarially: among the mistakes available, one of the two whose update leaves w at the widest angle to w*.