Linear Classification Models: From Binary to Multiclass

This lecture covers the transition from binary to multiclass classification problems, exploring the extension of linear classifiers like least-square classification and logistic regression to handle multiple classes. It delves into the challenges faced in multi-class scenarios, such as the lack of natural order between categories and the use of one-hot encodings. The application of these concepts is demonstrated through examples like predicting fetal states from cardiotocography data and thyroid disease classification. The lecture also compares different linear classifiers, including SVM, on datasets like UCI Iris and MNIST, showcasing their accuracies and training/testing times.