Bridging Scales Using Physically-Informed Machine Learning

Explore a comprehensive conference talk on physically-informed machine learning approaches for bridging scales in complex systems. Learn about the SPIDER framework, a novel methodology for inferring interpretable equivariant continuum models that has been validated across multiple experimental and numerical settings. Discover how this data-driven approach can derive coarse-grained descriptions from both continuum and discrete microscopic models. Examine two detailed case studies: subgrid-scale modeling of fluid turbulence and the inference of continuum models for molecular gases with repelling interactions. Gain insights into how machine learning can be integrated with physical principles to create more accurate and interpretable models for biological, chemical, and physical systems. Understand the theoretical foundations and practical applications of this emerging field that combines computational physics with advanced machine learning techniques.