Communicating neural network knowledge between agents in a simulated aerial reconnaissance system

Abstract

In order to maintain their performance in a dynamic environment, agents may be required to modify their learning behavior during run-time. If an agent utilizes a rule-based system for learning, new rules may be easily communicated to the agent in order to modify the way in which it learns. However, if an agent utilizes a connectionist-based system for learning, the way in which the agent learns typically remains static. This is due, in part, to a lack of research in communicating subsymbolic information between agents. In this paper, we present a framework for communicating neural network knowledge between agents in order to modify an agent’s learning and pattern classification behavior. This framework is applied to a simulated aerial reconnaissance system in order to show how the communication of neural network knowledge can help maintain the performance of agents tasked with recognizing images of mobile military objects.