Advanced Machine Learning

Aim

The course presents the analysis of several large classes of models widely used in advanced machine learning, such as state-space models, gaussian processes, hidden Markov models, Bayesian networks, and Markov random fields. Students will learn about the structure and learning of these models, when they are applicable, how to use them in practical machine learning applications, and how to correctly interpret the results. The models are mainly analyzed from a Bayesian perspective. After completing the course, the student should be able to:

• use the introduced model classes to accurately formulate and solve practical problems.
• learn the parameters and perform predictions in the presented models.
• devise algorithms to make an agent learn how to act optimally by interacting with the environment, and
• implement the models and learning methods in a programming language.

Content

​The course covers advanced methods in machine learning that allow for modelling complex phenomena and predicting the outcomes of these phenomena. The following topics are included in the course:

• probabilistic graphical models, such as Bayesian networks and Markov random fields,
• hidden Markov models including the forward-backward and Viterbi algorithms,
• reinforcement learning including the Q-learning and REINFORCE algorithms,
• Gaussian process for classification and regression.