Towards Robust Model Identification in Interactive Influence Diagrams

Modeling the perceived behaviors of other agents improves the performance of an agent in multiagent interactions. We utilize the language of interactive influence diagrams to model repeated interactions between the agents, and ascribe procedural models to other agents. Procedural models offer the benefit of understanding how others arrive at their behaviors. As model spaces are often bounded, the true models of others may not be present in the model space. In addition to considering the case when the true model is within the model space, we investigate the case when the true model may fall outside the space. We then seek to identify models that are relevant to the observed behaviors of others and show how the agent may learn to identify these models. We evaluate the performance of our methods in two repeated games and provide experimental results in support.