FRANCESCO BACCHIOCCHI

My PhD project focuses on the development of machine learning techniques applied to principal agent
problems. In these problems, a principal commits to performing a task, and an agent best
responds to this commitment. These problems are ubiquitous in game-theoretic settings, as they
model several real-world scenarios, such as contract design (where a principal aims to incentivize
agents to take costly, unobservable actions), Bayesian persuasion (where a sender strategically
discloses information to influence a receiver’s decision), and Stackelberg games (where the
principal commits to a strategy, and the agent best responds as a consequence). However, much
of the existing literature suffers from practical limitations, such as assuming the principal has
complete knowledge of both the environment and the agent’s features. My research aims to
develop machine learning algorithms that operate effectively, even in the absence of prior
knowledge about the environment. This requires the study of new online and reinforcement
learning techniques, enabling the principal to learn and adapt in real-time with the final goal of
steering the agent toward desirable equilibria. The proposed algorithms are designed to guarantee
an upper bound on the regret suffered during their execution while being efficiently implementable
in practice. By addressing these challenges, my research seeks to advance our understanding of
sequential decision-making in strategic environments characterized by partial or incomplete
information.