BIMSA >
BIMSA Digital Economy Lab Seminar
Multi-scale Modeling and Machine Learning for Discovering Hidden Laws in Complex Systems
Multi-scale Modeling and Machine Learning for Discovering Hidden Laws in Complex Systems
Organizers
Speaker
Time
Friday, March 28, 2025 3:00 PM - 4:00 PM
Venue
A3-2a-302
Online
Zoom 637 734 0280
(BIMSA)
Abstract
Deriving hidden laws from observational data and predicting the future state of systems is both fundamental and challenging. In fields such as chemical reactions, physical processes, and biological systems, these problems often involve highly complex and nonlinear dynamic behaviors. Multi-scale modeling can simultaneously consider system behaviors at different time and spatial scales, capturing interactions between different levels. Combined with machine learning algorithms, it can automatically identify patterns at different scales within a multi-scale framework, thus more efficiently and accurately revealing system dynamics when facing complex, large amounts of data. We apply machine learning methods to multi-scale modeling of chemical reactions, illustrating how multi-scale modeling can significantly reduce the computational cost of machine learning, and how machine learning algorithms can automatically perform model simplification in systems with time scale separation. We also developed a two-phase approach for learning interaction kernels in stochastic many-particle systems. Numerical experiments demonstrate excellent performance across various cases, including cubic, repulsion-attraction power-law, double-well potentials.