Classification

Binary Classification

1. Classification FramingInputs to labels: the decision rule, the boundary preview, and what classification promises.1–2 h
2. Scores and LogitsContinuous conviction behind hard labels: scores, thresholds, and the logit scale.1–2 h
3. Sigmoid and ProbabilitiesSquashing scores into probabilities: the sigmoid curve and its probabilistic reading.1–2 h
4. Logistic RegressionA linear score, squashed: the first probabilistic classifier, fit by loss.1–2 h
5. Decision BoundariesWhere the model changes its mind: boundary geometry for linear classifiers.1–2 h
6. Binary Cross-EntropyThe price of confident wrongness: log-loss for probabilistic labels.1–2 h
7. The PerceptronThe first learning rule: mistake-driven updates that move a linear boundary until data are separated.1–2 h
8. Generative Classifiers: Gaussian Discriminant AnalysisModel each class's inputs, then classify with Bayes' rule; shared covariance gives linear boundaries, separate gives quadratic.1–2 h

Multiclass Classification

1. Multiclass Scores and SoftmaxMany scores, one distribution: softmax turns a vector into probabilities that sum to one.1–2 h
2. Cross-Entropy LossOne loss for distributions: cross-entropy generalizes log-loss to many classes.1–2 h