Interpretable Machine Learning with an Eye for the Physics

Explore interpretable machine learning with a focus on physics in this 59-minute webinar presented by Harald Martens. Discover how to make sense of quantitative big data from multi-channel measurements of real-world complexity, such as hyperspectral or thermal video cameras. Learn about the challenges of converting overwhelming raw data streams into meaningful and reliable information through mathematical modeling. Compare the limitations of purely theory-driven modeling using classical differential equations and purely data-driven modeling using black box artificial neural networks. Delve into an alternative approach developed by the Big Data Cybernetics group at NTNU and NTNU spin-off company Idletechs AS. Examine the four levels of subspace modeling in this hybrid multivariate approach: theory-driven multivariate pre-processing, data-driven multivariate "chemometrics," traditional machine learning, and correction for alias errors. Gain insights into combining limited prior understanding with big data measurements to create more effective and interpretable machine learning models that consider the underlying physics of the systems being studied.