Uncertainty Quantification and Stochastic Optimal Control: Navigating Complexity in Forecasting and Decision-Making
Event details
| Date | 16.04.2025 |
| Hour | 16:15 |
| Speaker | Prof. Raul Tempone, KAUST - King Abdullah University Of Science And Technology, Arabie saoudite |
| Location | |
| Category | Conferences - Seminars |
| Event Language | English |
Abstract: In the era of data-driven decision-making, accurately quantifying and managing uncertainty is crucial across scientific and engineering disciplines. This talk introduces recent methodological advancements in Uncertainty Quantification (UQ) and Stochastic Optimal Control (SOC), highlighting novel computational techniques and their relevance in critical real-world scenarios.
In the first part, we discuss innovative UQ methods applied to complex systems, modeling the forecast error using Stochastic Differential Equations (SDEs). We present a data-driven approach that effectively quantifies uncertainties with concrete applications to renewable energy forecasting (wind and solar photovoltaic power). We also demonstrate a post-hoc UQ strategy for enhancing reliability in machine-learning-based biomedical image segmentation, specifically for estimating left ventricle volumes in cardiac MRI.
The second part of this talk focuses on recent methodological developments in SOC. Here, we present a framework for continuous-time stochastic optimal control under discrete-time partial observations, where control decisions integrate noisy measurements through a Bayesian update characterized by interlaced Hamilton–Jacobi–Bellman (HJB) equations with an infinite dimensional state.
Practical information
- General public
- Free
Organizer
- Prof. Fabio Nobile