Probability & Statistics Seminar

Upcoming sessions :

• Wednesday 28.10.2026, 15h30, room MNO 1.050
  Katrien Antonio (KU Leuven), Machine learning in an expectation-maximisation framework for nowcasting
  Abstract: Information is often only partially observable. In decision making, this may cause under or overestimation of underlying risk. Leveraging the available information to model the complete information is called nowcasting within the literature. In practical nowcasting applications, partial information is often caused by reporting delays. In this paper, we propose an expectation-maximisation framework that uses machine learning techniques to model both the occurrence as well as the reporting process of events. We allow for the inclusion of information specific to the occurrence and reporting periods as well as information related to the entity for which events occurred. Additionally, we demonstrate how deep learning techniques can be adapted for use in a nowcasting application. With simulation experiments, we show that we can effectively model both the occurrence and reporting of events when dealing with high-dimensional covariate information. In the presence of non-linear effects, we show that our methodology outperforms existing expectation-maximisation frameworks that rely on generalised linear models. We also show our ongoing research on using the developed nowcasting framework for modelling the claim dynamics of weather-related insurance claims.

• Thursday 03.12.2026, 13h30, room MNO 1.050
  Stéphane LOISEL (LIRSA Laboratory, Conservatoire National des Arts et Métiers, Paris), TBA
  Abstract: TBA

Past sessions :

• Thursday 10.09.2026, 13h30, room MNO 1.010
  Fabrice Baudoin (Aarhus University), Moment Estimates and Intermittency for the Stochastic Heat Equation on Fractals
  Abstract: We study the parabolic Anderson model, or stochastic heat equation with multiplicative space-time white noise, on bounded fractal spaces, with particular emphasis on the Sierpiński gasket equipped with the Kigami Laplacian and Dirichlet boundary conditions. A central question is how the geometry of the underlying space influences the large-time behavior of the solution. Using heat kernel estimates, Wiener chaos expansions, and hypercontractivity, we derive upper bounds for the growth of moments of the solution and obtain matching lower bounds in the second moment. These estimates quantify the competition between the dissipativity of the Dirichlet heat semigroup and the excitability induced by the multiplicative noise. They also provide evidence for intermittency phenomena on fractals, where high peaks of the solution carry most of the mass at large times. The results reveal the role of the Hausdorff and walk dimensions in determining the Lyapunov-type exponents governing moment growth.