Check your understanding
Quiz
You have worked through the sharp end of supervised learning: maximum-margin classifiers and the kernel trick, regularized discriminant analysis, Gaussian processes that predict with honest uncertainty, stacking several models into a Super Learner, Bayesian optimization for expensive tuning, approximate nearest neighbors at scale, and the discipline that ties it together, tuning and evaluating one pipeline without fooling yourself. This quiz checks what stuck. The last two steps are live R you can run.
Question 1 of 10
The maximum margin
Two clean classes can be separated by many different straight lines, all with zero training errors. Why does a support vector machine choose the one with the widest margin?
Correct: equal training accuracy cannot separate the candidates, but the widest-margin boundary is the most robust to unseen points that drift toward it.
The margin is precisely the quantity that distinguishes otherwise-equal boundaries on new data.