| Exploration in Multi-Agent Systems |
|
Organization responsible: UvA People involved: Project description: We propose to undergo theoretical and practical research to extend current methods for addressing the exploration-exploitation trade-off in sequential decision making problems to multi-agent systems. While the focus is on practical Bayesian methods for distributed exploration, there is a possibility for more theoretical work in bounds for simple algorithms that perform distributed exploration. Taking crisis management as an example, there is usually partial information available about the crisis from multiple sources. The problems of information fusion and uncertainty in state estimation have been addressed to a large extent, but the problem of choosing between further exploration or taking a decisive action has not received much attention. The related topic of performing exploration such as to maximize the probability of obtaining useful information, especially in multi-agent systems, has not been explored either. The proposed research shall bring together expertise in information fusion, partially observable multi-agent problems and exploration. Publications: |