Physics-guided Learning in Control

We live in an era of information in which traditional modeling methods are being substituted by data-driven solutions. The motivation towards this course comes from the fact that data is abundantly found everywhere. In addition, the latest advances in machine learning allow us to extract patterns from data in a way that was not computationally feasible before. Although these data-driven solutions are very promising and offer very accurate solutions for complex systems, the end result is rarely understandable and often viewed as a "black-box". With this project, we want to explore the possibilities of enhancing first-principle mathematical models (e.g. Euler-Lagrange, port-Hamiltonian, thermodynamic models) with recent regression methods, in an attempt to obtain insightful representations of the underlying dynamics of complex systems.