Modeling of Agent Decisions Using Conditional Generative Adversarial Networks

In this paper, we investigate the use of Generative Adversarial Networks (GAN) to model agent behavior in agent-based models. We hereby focus on use cases in which an agent's decision-making process may only be modeled from data, but it is infeasible to be modeled causally. In these situations, meta-models are often the only way to quantitatively parameterize the agent-based model. However, methods that capture not only deterministic relationships but also stochastic uncertainty are rare. Since GANs are well known for their property to generate pseudo-random-numbers for complex distributions, we explore pros and cons of this strategy for modeling a delay-process in a large-scale agent-based SARS-CoV-2 simulation model.