Learning in Heterogeneous Environments: a Case Study

We propose a learning mechanism suitable for using in a heterogeneous environment, where interconnected systems present simultaneously a competitive and cooperative behaviour. By interconnected systems we understand systems that need to interact with each other and then act either in an individual as well as a global way: each of the systems observes itself and the environment, and acts according to its unique perspective; but a system can also need the help of others to complete its own tasks. In our case, systems are represented by agents, since we are studying environments where autonomy and pro-activity are essential features. The learning mechanism we propose here can be classified as a multi-agent learning mechanism, not only because there are multiple agents learning concurrently in the same environment but also because it allows them to understand how to improve their own performance without degrading the performance of the other agents. We tested our learning mechanism over the Disruption Management in Airline Operations Control Center application domain, where interconnected agents (namely the Aircraft Manager, the Passenger Manager and the Crew Manager) need to combine their expertise. The results show that our learning mechanism provides a good performance to the agents that operate simultaneously in cooperative and competitive situations.