Project ideas for Part II / Part III / MPhil ACS
Incorporation of expert knowledge in Reinforcement Learning (RL)
RL allows agents to learn waht actions to take given their experince of inteacting with the environment, and the reward and punishment associated with these experinces. Incorporating experts knowledge of the domain in the learning process helps improving the performance of RL agents. However, domain experts knowledge can be self-conflicting. In order to use this knowldge as heuristics incorporated into RL, the agents have to be able to reason about conflicting knowledge of domain experts. The aim of this project is to equip agents with such reasoning capability.
The ideal candidate for this project will have great programming skills (preferred programming language: Java). Good knowledge of reinforcement learning is required too. Familiarity with Robot Soccer simulation platforms (FIRA/RoboCup) is a plus.
Coordination in Multi-Agent Systems (MAS)
Learning to coorinate in MAS to achieve a common goal is a challenging task, in part due to the large possible joint actions between agents. Agents need to resolve their conflicts to reach agreement on wht to do. To resolve conflicts, arguments in favour and against specific action-agent pair can be assessed in search of the strongest arguments recommending which action should be taken by each agent. The aim of this project is to improve performance of reinfocemenet learning agents in a RoboCup Takeaway game by using argumentation.
The ideal candidate for this project will have great programming skills (preferred programming language: Java). Good knowledge of reinforcement learning is required too. Familiarity with Robot Soccer simulation platforms (FIRA/RoboCup) is a plus.