Learning Problem Solving in Cooperative Multi-agent Systems

Contents

The work presented here deals with ways to improve problem solving control skills of cooperative agents. We propose situation-specific learning as a way to learn cooperative control knowledge in complex multi-agent systems. We demonstrate the power of situation-specific learning in the context of dynamic choice of a coordination strategy based on the problem solving situation. We present a leaming system, called COLLAGE, that uses meta-level information in the form of abstract characterization of the coordination problem instance to leam to choose the appropriate coordination strategy from among a class of strategies.