Demo 04

Classic Machine Learning: How do machines learn decision boundaries from data?

Classic machine learning shifted from hand-written rules to data-driven methods. This chapter shows the relationship among examples, model complexity, and decision boundaries.

Teaching interaction

Decision Boundary Explorer

How do machines learn classification boundaries from data?

  • Positive examples
  • Negative examples
  • Draggable negative-class outlier

A linear model learns a simple dividing line

A linear boundary is stable and interpretable, but its expressive power is limited.

Moving the outlier now changes the displayed boundary. The flexible modes react more strongly, illustrating sensitivity rather than a real training run.

Learning goals
  • Understand how classic machine learning learns a decision boundary from examples.
  • Compare the inductive biases of linear, nonlinear, and overfit boundaries.
  • Observe how an outlier can affect a model boundary and its generalization risk.
Simplification note

This demo changes preset boundaries according to the outlier position and each mode's teaching sensitivity to illustrate inductive bias. It does not train a real model or represent actual classifier performance.

Observation guide

A model learns a boundary from examples instead of memorizing rules

The earlier problem was that rules could not cover every situation. Classic machine learning learned patterns from data, but it did not automatically solve feature design, generalization, bias, or overfitting. Its follow-on influence includes SVMs, tree models, ensemble learning, and deep learning.

References

Concept Check

Test The Core Intuition With One Question

This check supports reflection and never blocks the next chapter. Records stay on this device.

Why can moving one outlier change the decision boundary in the demo?