Reduced-order models offer computationally efficient approximations of complex systems, enabling multi-query tasks in design and optimisation with low cost and sufficient accuracy. Data-driven strategies are particularly appealing when underlying models are inaccessible or too expensive to evaluate, and recent advances in AI-based architectures have naturally entered this space. However, these architectures still face challenges when confronted with systems exhibiting variable dynamics, bifurcations, or chaotic behaviour. In this talk, we present a shift in perspective that unifies complex dynamical systems with nonintrusive, data-driven reduced-order modelling approaches, thereby broadening the range of applications that can be addressed effectively.
By Matthew Juniper
By Sergey Gavrilyuk
By David Lannes
