Advanced Learning

Instructor: Francis Bach.

ENS, September-December.

Course contents:

Typology of learning problems (supervised vs. unsupervised). Statistical model for binary classification: generative vs. discriminative approaches. Classical algorithms: parametric methods, perceptron, partitioning methods. Performance criteria: classification error, ROC curve, AUC. Risk convexification: boosting and SVM algorithms. Combinatorial complexity measures, geometric metrics. Model selection and regularization. Consistency theorems and convergence speeds.