DataScience@BI seminar with Martin Bergerhausen
DataScience@BI invites Martin Bergerhausen, PhD candidate, University of Mannheim, to give a talk on ‘Neural Stochastic Volterra Equations’.
- Starts:12:00, 5 May 2026
- Ends:13:00, 5 May 2026
- Location:BI - campus Oslo, B3 inner area - next to meeting room B3i-108 or Zoom
- Contact:Siri Johnsen (siri.johnsen@bi.no)
Abstract:
Stochastic Volterra equations (SVEs) extend classical SDEs by incorporating memory effects and non-Markovian dynamics through kernels. While these features are essential for modelling complex systems with inter-temporal dependencies, the loss of the Markov property poses significant challenges for classical numerical schemes and by limiting the mathematical toolbox.
In this talk, we address the problem of learning SVE dynamics directly from sample paths. We introduce Neural SVEs, a framework where the underlying drift, diffusion, and kernel operators are parametrized by neural networks. By embedding the Volterra integral structure into the architecture, we ensure that the model respects the inherent causality and memory of the system.
We discuss the mathematical advantages of this physics-informed approach over similar methods. Numerical experiments demonstrate that Neural SVEs provide superior generalization in scenarios where temporal correlations are dominant, offering a robust path toward hybrid modelling in stochastic analysis.
The DataScience@BI seminar is organised by the Department of Data Science and Analytics, BI Norwegian Business School.