Schaefer: Learning from Data - WiSe 24/25

This course provides an introduction into machine learning theory including mathematical foundation. It covers the following areas:

• Is statistical or machine learning possible? (Outlook Vapnik-Chervonenkis theory, VC Dimension and Limit)
• Statistical/Machine learning: Supervised, unsupervised, reinforcement learning, classification, regression
• Probabilistic and non-probabilistic methods
• Different models:
  • Linear model
  • Kernel methods
  • Support Vector Machines
• Bayes methods
  • Gaussian processes
  • MCMC and particle filters
• Neural networks
• Genetic algorithms
• Model selection
  • Overfitting and regularization
  • Bias-Variance tradeoff, error and noise
• Training, testing, validation o Curse of dimensionality
• Vapnik-Chervonenki theory, Hoeffding’s lemma, VC dimension and VC inequality
• Structural Risk Minimization (SRM), Occham’s Razor
• Overview of selected application fields
• Consequences of statistical learning theory
• Philosophic implications (Are data more important than theories? Where are the limits of statistical learning theory?
• Ethical implications for society, principal problems of algorithmic decisions (Information privacy, individual freedom, checks-and balances of citizens, corporations, states etc.)
• Lectures are accompanied by practical exercises with respect to theories and concepts, supported by the use of appropriate software