Student should be able to understand and describe::
- The concept of learning from data in contrast to rule-based systems.
- The differences between the various machine learning paradigms (supervised learning, unsupervised learning, reinforcement learning, active learning).
- The statistical foundations of machine learning (predictor, loss function, statistical risk, Bayes optimal prediction, Bayes risk, bias-variance decomposition, overfitting and underfitting, consistency, regularization, stability, model validation and model selection).
- The basic linear models for regression and classification (linear regression and linear classification, logistic regression, Support Vector Machines).
- The theory of kernel functions and the implementation of linear models in kernel spaces.
- The notion of nonparametric learning, its connection to consistency, the curse of dimensionality, and the basic nonparametric algorithms (e.g., k-NN, tree predictors, universal kernels).
- Probabilistic models (e.g., Naive Bayes classifiers, graphical models, Gaussian mixture models).
- Ensemble learning methods (bagging, boosting, stacking) and their effect on the bias-variance dilemma.
- Feed-forward neural networks, expressivity versus network size, hardness of training, backpropagation algorithm.
- Basics of online learning and reinforcement learning (e.g., online gradient descent, multi-armed bandits, Markov decision processes with finite and discounted horizon, model-based and model-free RL algorithms).
- Basic clustering algorithms (e.g., k-means and k-means++, DBSCAN, hierarchical clustering, correlation clustering).
- Model validation and selection (e.g., cross-validation, nested cross-validation).
- Basic techniques for dimensionality reduction (e.g., LDA, PCS, ISOMAP, LLE).