**Advisors** Michel Baes, Patrick Cheridito, Florian Rossmannek and Philipp Zimmermann

**Time** Wed 3:15 – 5:00 pm

**Main Reference**

**Additional Reading**

**Presentation Tips**

**Meetings**
**Sept 18** Organization of the seminar
**Sept 25**
What is learning? ERM, PAC, agnostic PAC, uniform convergence

(UML 2–4; HDP 2.6; advisor: Florian)
**Oct 2** Examples of hypothesis classes, halfspaces, regression, no-free-lunch theorem

(UML 5, 9; advisor: Florian)
**Oct 9** A combinatorial approach to learnability, VC-dimension, fundamental theorem of statistical learning
(UML 6; HDP 8.3.1–8.3.3; advisor: Michel)
**Oct 16** Hypothesis dependent sample sizes, non-uniform learnability, PAC-Bayes (UML 7, 31; advisor: Michel)
**Oct 23**
Convex learning problems, RLM, stability (UML 12–13; advisor: Philipp)
**Oct 30** Stochastic gradient descent (UML 14; advisor: Philipp)
**Nov 6** Kernel methods, multiclass, ranking, complex prediction problems (UML 16–17; advisor: Michel)
**Nov 13** Multiclass learnability, Natarajan dimension, fundamental theorem for multiclass problems (UML 29; advisor: Michel)
**Nov 20** Feedforward neural networks (UML 20; advisor: Florian)
**Nov 27** Generative models, MLE, EM, Bayesian reasoning (UML 24; advisor: Florian)
**Dec 4** Dimensionality reduction, PCA, sparse recovery, compressed sensing

(UML 23; HDP 10.1–10.3; advisor: Philipp)
**Dec 11** Online learning, Littlestone dimension, online gradient descent (UML 21; advisor: Philipp)
**Dec 18** More precise sample size estimates, Rademacher complexities, covering numbers,
Dudley's inequality (UML 26–27; HDP 8.1–8.4; advisor: Florian)