Seminario Machine Learning
Statistical Methods for Digital Health: To Infinity and Beyond
Ponente: Marcos Matabuena (Mohamed bin Zayed University of Artificial Intelligence (MBZUAI))Fecha: lunes 22 de junio de 2026 - 12:00Lugar: Aula Gris 2, ICMAT
Resumen:
Digital health technologies collect data continuously over time, opening new opportunities to personalize healthcare and enabling a new generation of biological discoveries. In this context, a natural analytical framework is to work with functional spaces, since we often observe multiple curves recorded at different time scales; for example, daily glucose trajectories collected over several days. However, in real-world studies, time series are not always directly comparable: individuals exercise at different times, eat at different moments, and are observed under conditions that are difficult to standardize. For this reason, in many cases, a more robust biomarker is not the observed temporal trajectory itself, but rather the probability distribution associated with the time series, which can be viewed as a random statistical object in a metric space. The goal of this talk is to introduce the main analytical challenges that arise in this new era of digital health, a field that I refer to as digital biostatistics. I will discuss new statistical models developed in my recent work, with particular emphasis on random effects in metric spaces and conformal prediction. These models will be illustrated through relevant applications in digital health.
