Perspectives Talk III - Industry: Physics-Grounded Machine Learning

Machine learning has demonstrated impressive achievements for a wide range of applications, but many systems are unable to provide a high level of reliability and trustworthiness in the results. This is especially important for industrial and safety-critical systems, where a higher level of assurance in the results are essential. This talk will offer a perspective on the emerging area of physics-grounded machine learning for the design, optimization, and control of real-world engineering systems. Such a framework can enforce physical principles and constraints, while still leveraging the power of data-driven machine learning techniques. Several industrial applications will be covered to demonstrate the benefits of this framework including radar-based imaging, which leverages the physics of wave propagation; airflow sensing and optimization, which is governed by the Navier-Stokes equation; as well as a range of additional applications that can be modeled as a dynamical system or through geometric constraints.