Research Interests

My main research focuses on sequential decision-making processes. My interests span the areas of biostatistics, Bayesian methods, statistical reinforcement learning & multi-armed bandits, and modern applications based on adaptive decision making, which include Dynamic Treatment Regimes (DRTs), mobile-Health (mHealth), and other adaptive experimentations such as educational trials.
More specifically, my focus is on using multi-armed bandit strategies for better designing novel trials such as adaptive clinical trials and micro-randomized trials. Additional interests are given to the emergent area of mobile-Health and how to optimally design mobile apps for guiding users’ behavior and improving their health outcomes. The design of such applications involves both the design of the trial itself and the design of the (bandit) algorithm. As such kind of experiments may have a dual goal of improving outcomes for users enrolled into the experiment and learning about the effectiveness of interventions, estimation and inference in adaptively collected data play a central role in my research arena.

During my Master’s I focused on Bayesian Statistics, specifically Markov Chain Monte Carlo methods, and Deep Learning Algorithms, both with (spatio-)temporal applications.

After starting my PhD, I got really interested in Reinforcement Learning (RL) and its statistical properties in terms of inference and causality, particularly relevant in bio-medical settings. My first approach was to carefully review all the RL algorithm applied in modern biostatistic, identifying, first, all the health-related applications, and then the relative methodologies.

Currently, I’m focusing on problems and challenges arising in some of these specific applications and the related statistical approach evaluation. More specifically, my attention is now on:

• MABs with non-stationarity settings, typical of real-world mHealth problems where habituation is a main challenge;
• Non-stationarity and habituation modeling in complex multi-factorial designs;
• Statistical inference in MABs online adaptive experiments, where the primary interest is generally on maximising an expected cumulative reward;
• Design of adaptive randomized trials.

I have also worked for a couple of years in a clinical foundation, collaborating with clinicians and healthcare researchers. I think that statistics and machine learning (ML) can bring a lot in advancing the health-related research and, reversely, that many of the interesting challenges and problems in statistics and ML araise from real-worls applications.