| Objectives |
The students learn basic concepts of machine learning (ML) like supervised, unsupervised, semi-supervised methods, and reinforcement learning. Students acquire knowledge on basic ML concepts like parametrized and unparametrized models, model selection, hyperparameter selection, feature selection, overfitting and underfitting (bias and variance), objective and loss functions with their risks, and more. Evaluation concepts for methods like test set and cross-validation will be known by the students.
In supervised learning, basic machine leaning approaches and tools like k-nearest neighbors, decision tree approaches like naïve Bayes classifier, random forest, kernel methods like support vector machines, boosting methods like gradient boosting, time series prediction methods, and neural networks are covered. Furthermore, students are familiarized with the most important regularization methods.
In unsupervised learning, the concepts of recoding methods vs. generative methods are presented. Different unsupervised objectives like energy, entropy and information maximization, independence, cluster separation, and in particular maximum likelihood with the EM algorithm are taught, as well as basic techniques such as PCA, ICA, factor analysis, projection methods (Isomap, LLE, multidimensional scaling and Samon maps, t-SNE), clustering methods, mixture methods, biclustering methods, Boltzmann machines, and hidden Markov Models.
In reinforcement learning, the students learn Markov processes, Markov decision processes (MDPs), partially observable MDPs, the action-value function (Q-function), the value function, the Bellman equations, Monte Carlo estimates, Q-learning, (expected) SARSA, eligibility traces, policy gradient methods like REINFORCE, exploration techniques, and regularization techniques.
|
