Raphaël Huser

Research Interests

Raphaël Huser’s research primarily focuses on statistics of extreme events and risk assessment, including developing specialized statistical models with appealing statistical properties. Additionally, he studies efficient machine learning methods designed for massive datasets from complex spatio-temporal processes.

Huser’s novel methodological contributions are motivated and inspired by a wide variety of real data applications, which include the modeling of natural hazards in climate and earth sciences (e.g., heavy rainfall, heat waves, extreme sea surface temperatures, strong wind gusts, devastating landslides), the assessment of financial risk (e.g., turbulence in stock markets or cryptomarkets), and the characterization of brain signals during extreme stimuli (e.g., epileptic seizures).

Beyond creating new models with interesting statistical properties, one crucial aspect is fitting these complex models to big data. A critical area of Huser's current research is developing general-purpose, likelihood-free, fast and statistically efficient neural Bayes estimators. Being deeply anchored in statistical decision theory and relying on advanced deep-learning techniques, which makes them attractive both from a theoretical and a computational perspective, these estimators truly provide a paradigm shift challenging traditional statistical inference techniques for complex models with intractable likelihoods.

Huser, with collaborators, is contributing extensively to the early development of such estimators and their application to spatial (e.g., extreme) and multivariate models.