Abstract

Targeted interventions in games present a challenging problem due to the asymmetric information available to the regulator and the agents. This note addresses the problem of steering the actions of self-interested agents in quadratic network games towards a target action profile. A common starting point in the literature assumes prior knowledge of utility functions and/or network parameters. The goal of the results presented here is to remove this assumption and address scenarios where such a priori knowledge is unavailable. To this end, we design a data-driven dynamic intervention mechanism that relies solely on historical observations of agent actions and interventions. Additionally, we modify this mechanism to limit the amount of interventions, thereby considering budget constraints. Analytical convergence guarantees are provided for both mechanisms, and a numerical case study further demonstrates their effectiveness.