Learning Equilibria in Strategic Asymmetric Environments

Abstract:
Every day, a very large number of users interact with digital platforms such as online advertising platforms or crowdsourcing marketplaces. These interactions are characterized by a strong asymmetry, not only in objectives, but also in the information available to users and to the centralized platform. Game theory has traditionally captured the nature of such interactions through the notion of Stackelberg equilibrium, where the platform acts as a leader by committing to a certain strategy and users, acting as followers, respond based on their private information and objectives. However, most existing work mainly focuses on static environments in which the leader knows everything about the followers beforehand. In practice, this assumption rarely holds. Platforms must set prices, incentives, or information policies under substantial uncertainty. In this talk, we show how this limitation can be overcome by studying how Stackelberg equilibria can be learned through repeated interactions in different settings at the core of today’s digital economies, using data-driven, no-regret, and sample-efficient techniques, even when users’ preferences are initially unknown.

Biography:
Francesco Bacchiocchi is an ELLIS PhD student in Computer Science at Politecnico di Milano, supervised by Prof. Nicola Gatti and Prof. Jiarui Gan at the University of Oxford. He received both the M.Sc. and the B.Sc. in Mathematical Engineering from Politecnico di Milano. His research lies at the intersection of algorithmic game theory and online learning, with the goal of understanding how strategic agents behave in dynamic environments. His work has been published at top ML venues, receiving oral and spotlight presentations.