Abstract
The notion of data-driven modelling has been become practically synonymous with the use of increasingly complex neural network architectures to explore physical systems. This modelling approach promises to learn the system dynamics purely from observed trajectory data, returning an input-output relationship in the form of a convoluted network. This is to be contrasted with the conventional approach of applied science that seeks to understand the governing principles and derive a compact, interpretable representation of the core system behavior. Each approach claims benefits over the other, but systematic and fair in-situ comparisons on physical systems are rare.
In this talk, I discuss how data-driven modelling and control based on contemporary nonlinear dynamical systems methods compare with contemporary neural network-based methods in structural dynamics and soft robotics.
Imperiathe speaker
George Haller is a Professor of Mechanical Engineering at ETH Z眉rich, where he holds the Chair in Nonlinear Dynamics. His prior appointments include tenured faculty positions at Brown, McGill and MIT. He also served as the first director of Morgan Stanley鈥檚 mathematical modeling center. Professor Haller is a former Sloan Fellow, Thomas Hughes Young Investigator (ASME) and School of Engineering Distinguished Professor (McGill), as well member of the Hungarian Academy of Science and the Academia Europaea. He serves as associate editor at the Journal of Applied Mechanics, feature editor at Nonlinear Dynamics and senior editor at the Journal of Nonlinear Science. He is an elected fellow of SIAM, APS and ASME, the recipient of the 2023 Stanley Corrsin Award of the APS and the 2025 Lyapunov Award of the ASME.
This event is part of the hosted at the Department of Aeronautics.