Approximate game theoretic analysis for large simulation-based games
Bryce Wiedenbeck · Adaptive Agents and Multi-Agents Systems · 2014
1. SIMULATION-BASED GAME THEORY Multi-agent systems research raises many interesting questions about the actions of autonomous, interacting decisionmakers. Game theory helps to answer some of those questions by providing a mathematical language for representing and reasoning about the incentives of self-interested autonomous agents. As is often the case with mathematical modeling, creating tractable game-theoretic models frequently requires abstracting away many aspects of the system under investigation. Simulation-based game theory constructs richer game models using agent-based simulation, but still suers from tradeos between tractability and accuracy. My research agenda focuses rst on improving this tradeo by developing techniques for simulation-based game theory to conduct better analysis with limited simulation data and second on applying these techniques to understand complex interactions. My thesis will present novel methods that improve the tools available for simulation-based game theory and demonstrate the application of these tools to uncover new results in multi-agent systems of interest. The fundamental building block of a simulation-based game model is the environment simulator. The simulator takes as input a strategy for each agent, simulates those strategies interacting, and outputs a payo