Interactive dynamic influence diagrams modeling communication

Abstract

Communication embodied the social characteristic of multi-agent systems. Agents could benefit from exchanging information. When communication incurs a cost, whether to communicate or not also becomes a decision to make. In this paper, we study communication decision problems with an extension framework of interactive dynamic influence diagrams (I-DIDs), we assume a communication sub-stage where all communication complete before deciding the regular action. In our framework, agent at higher level has a choice of performing a communication action just after the previous action finishes and before the next action is chosen. The purpose of communication is for higher level agent to send its current observation (real or deceptive) to other agent at lower level. Agent initiates communication so as to optimize the benefit it obtains as the result of the interaction. An example problem is studied under this framework. From this example we can see the impact communication policies have on the overall rewards of agents. © 2012 Springer-Verlag Berlin Heidelberg.