Scientific Machine Learning for Science and Engineering

Description

Many problems in sciences, and engineering share similar challenges from the methodological perspective. They are typically characterised by high-dimensional state and parameter spaces, nonlinear dynamics and heterogeneous multi-modal and partly high-frequency state and condition observations, together with possibly noisy and limited data. The physics-based models are usually computationally very expensive and may not be fully representative of the underlying processes, since those may not be even completely understood. Machine learning approaches, on the other hand, have demonstrated substantial progress and potential in various disciplines, but they also face several limitations, including limited interpretability and extrapolation capability. The general challenges in sciences and engineering include that the collected data is not fully representative of all possible conditions and is often lacking labels that are required for training or high levels of imbalance in the label space. While the challenges in different disciplines in sciences, and engineering appear very different, innovative Scientific ML algorithms that span across domains enable fast, efficient and robust solutions to these problems.

Period	11 Feb 2025
Event title	The Applied Machine Learning Days (AMLD) EPFL 2025
Event type	Conference
Location	Lausanne, SwitzerlandShow on map
Degree of Recognition	International