Programme Outlines and Overviews

The quest to fully realize the potential of Artificial Intelligence (AI), requires autonomous systems that can learn to make good decisions by interacting with their environment. Reinforcement learning is a paradigm that meets these requirements, and can be applied to various tasks, including game-playing, healthcare, economics, and robotics. This course gives a solid introduction to reinforcement learning with its core approaches and challenges, and is structured around several lectures, assignments, and a project. The course includes the following elements: • Markov Decision Processes (MDPs) • Model-based and model-free prediction and control • On-policy and off-policy methods • Monte Carlo, Temporal Difference, Policy-Gradient, and Actor-Critic methods • The exploration versus exploitation trade-off, including regret • The bias versus variance trade-off, including stability • Function approximation, including Deep Reinforcement Learning • Imitation Learning and Reinforcement Learning with multi-agent interactions