Every model, drawn live.

Least squares through your points, from a straight line to a wiggly polynomial.

Predict the average of the k closest points. No equation, just neighbours.

Split the x-axis into steps and predict the mean inside each one.

Add small trees one at a time, each one fixing what the others got wrong.

A linear score squashed into a probability, trained by gradient descent.

The original neuron. Nudge the line every time it gets a point wrong.

One Gaussian per class, all sharing the same shape, which gives linear boundaries.

Every class gets its own Gaussian shape, so boundaries become curves.

Assume the features are independent within each class, then apply Bayes' rule.

Classify a point by a vote among its k closest neighbours.

A flowchart of yes/no questions that carves the plane into boxes.

Many slightly different trees vote, and the noise averages out.

Weak learners in sequence, each focusing on the points the last one missed.

The widest possible street between classes, bent by a kernel when needed.

Layers of neurons learn their own features, trained live with backpropagation.

Drop k centres, assign points, move centres, repeat until nothing changes.

Soft clustering with stretched ellipses, fitted by expectation-maximisation.

Clusters are dense regions. Any shape works, and outliers are left out as noise.

Merge the two closest groups, again and again, and read the clusters off a dendrogram.

Find the directions where the data spreads out the most.