Francesca Parise – School of Electrical and Computer Engineering, Cornell University, United States
Understanding the role of network interactions is fundamental for improving efficiency, resilience and welfare in many socio-economic settings. The large size of these systems (e.g., involving billions of users in the case of social platforms) however introduces some challenges from the perspective of a planner that aims at regulating interactions. In fact in many cases, collecting exact network data is either too costly or impossible due to privacy concerns. In these cases however it might be feasible for the planner to collect statistical information about agents’ interactions that can be used to infer a random graph model. A key question is then whether knowledge of such a random graph model is sufficient to infer relevant features of the realized network or to control a dynamical process evolving over it. This question has been addressed in a number of recent works by focusing on network games with unique Nash equilibrium. Yet, many relevant networks may exhibit multiple-equilibria. In this talk, we will present novel results in this direction. Specifically, we will present a convergence theory for graphon games with multiple equilibria and algorithms for learning in time-varying settings.