Given by Konstantin Tretyakov
Brief summary: Recap on algebra and geometry. Maximal margin classifiers. Reformulation as a quadratic programming problem. Primal and dual forms. SVM as an example of a regularized learning problem. Hinge loss as an example of a surrogate loss function.
Cristianini and Shawe-Taylor: An Introduction to Support Vector Machines pages 93 - 112
Schölkopf and Smola: Learning with Kernels pages 189 - 215
Two ways to minimise objective function hard and soft clustering. Soft clustering as expectation-maximisation. Robust Gaussian mixture models. Mixtures of Bernoulli distributions. Logistic regression and linear discriminant analysis.