Table of Contents

🎓 Intended learning outcomes

At the end of this lesson, students are expected to:

• Motivate classification from a geometric perspective
• Derive the perceptron from basic principles and apply it
• Understand and motivate margin classifiers
• Formulate a constrained optimization problem for margin classifiers
• Define equality constraints, inequality constraints, and feasible region
• Describe the need and implementation of slack variables
• Derive a soft-margin linear SVM formulation from two perspectives
• Motivate and identify regularization terms in the average loss
• Understand and implement subgradients, including the Pegasos algorithm
• Be able to define and explain square loss, logistic loss, hinge loss, and zero-one loss

📐 A geometric perspective ★☆☆