The increasing availability of large and complex ground displacement datasets creates new opportunities for natural hazard monitoring, but also poses significant methodological challenges arising from the high dimensionality of the data and from the presence of noise, broadly understood as partial observation and contamination in the recorded signals.

My research develops nonparametric statistical methods for reliable inference in high-dimensional and noisy data, motivated by these challenges. Ground displacement measurements can naturally be represented as realizations of underlying continuous physical processes, for which functional data analysis provides a principled framework. Complementarily, conformal prediction enables finite-sample valid inference without relying on strong modeling assumptions. The combination of these perspectives yields robust and interpretable methods for monitoring ground displacement and natural hazards with explicit control of false alarms. Recent work investigates conformal methods that remain valid and informative under data contamination, addressing classification and scalar regression as initial steps toward more complex settings.

Future research will extend conformal inference to regression problems with contaminated or partially observed responses, with the goal of improving the reliability and power of monitoring and forecasting tools for ground displacement data. At the same time, the nonparametric nature of these methods makes them applicable more broadly to high-dimensional and noisy data in other domains.

Contatto:
andrea1.manzoni@polimi.it