|
|
|
|
|
Objectives
Understanding recent developments in single- and multi-agent decision-theoretic planning.
Contents
This course addresses the question: how can we design effective control policies for single- and multi-agent systems? The focus is on decision-theoretic methods, wherein optimal or approximately optimal solutions are discovered via planning or learning. In the single-agent case, topics such as exploration, function approximation, partial observability, and policy search will be covered. In the multi-agent case, topics such as coordination, game theory, coevolution, and multi-agent learning will be covered. The course focuses on reading and discussing recent research papers on these subjects. Each student is required to present one such paper to the class and also to complete a programming/research project extending an existing algorithm or applying it in a novel setting.
Registration at
Registration for courses is mandatory, but will be done by the Education Service Centre for the 1st year MSc students for courses of the first semester. See also
http://www.student.uva.nl and choose your master and then 'New procedure 'Registration for courses Faculty of Science'.
Format
Lectures and lab course.
Assessment
Weekly paper responses, class participation, a paper presentation, and a research project report.
|
|
|
|
|