External Coordination Media in Capacity-Constrained Multiagent Systems

Our goal is to understand limitations of simplicity of knowledge structures and reasoning processes in Multiagent Systems. Therefore, we propose a framework that integrates external storage media and a capacity-constrained Multiagent System. Agents can store knowledge internally and on external storage media located in an environment. In some cases, agents either have to forget or to store knowledge externally due to limited internal memory. We define notions of suitable knowledge and let agents learn in the context of an iterative partitioning task problem. Using basic knowledge structures and a simple k-Nearest-Neighbor approach, we evaluate the role of limited internal memory in conjunction with strategic positioning of storage media and knowledge items. We also investigate dependencies between communication, internal memory size, frequently changing settings, and externally stored items. The results show that externally and strategically stored simple knowledge can support agents' reasoning processes. The approach is robust even in highly dynamic settings with small memory and without inter-agent communication.