Game-Family Learning for Simulation-Based Games

My Ph.D. research introduces an approach to learning families of parametrically related symmetric game instances in both normalform and Bayesian settings. This approach eliminates the need to model each game instance separately, improving data efficiency and enabling broader exploration of the parameter space. Gamefamily models support more comprehensive empirical mechanism design, and facilitate iterative generation of piecewise best-response strategies in the Bayesian setting. Overall, my work aims to expand model expressiveness while prioritizing compactness and data efficiency in simulation-based game environments, promoting diverse real-world insights.