Reinforcement Learning deals with how an agent decides on actions to take in an environment, while intending to maximise some notion of a reward. Deep neural network models now allow Reinforcement Learning methods to solve complex problems end-to-end. The module will cover decision and control theory, exploration, Q-learning and policy gradients, hierarchical Reinforcement Learning, Markov decision processes, model-based Reinforcement Learning, multi-agent Reinforcement Learning, planning and navigation.
Upon completion of the module the student will be able to:
- summarise, implement and critically assess basic as well as state-of-the-art techniques to solve common Reinforcement Learning and Planning problems;
- formulate and interpret the mathematical theory underlying common Reinforcement Learning and Planning approaches;
- apply a number of devices necessary to build practical solution to Reinforcement Learning and Planning problems;
- discuss problems and approaches at the cutting edge of Reinforcement Learning and Planning research, from an ML and AI perspective.