Efficient Reinforcement Learning in Multi-Agent Systems
Abstract
This project will investigate efficient methods for reinforcement learning in multi-agent systems, e.g. reward shaping, curriculum learning and hierarchical reinforcement learning. The intended application for the techniques is simulation-based training systems, which adopt a gamified approach. The project will thus use game environments and deep learning frameworks for training of intelligent agents. The project is connected to a research project within the national aeronautical research program,NFFP7.