Physics-based models and purely data-driven machine learning models each have significant benefits and limitations. A new pathway, differentiable modelling, combines the two methods and pushes the boundary of physics-informed machine learning. I will use some examples from water resource research to demonstrate advantages of this combined approach where we have mitigated the limitations of each method, more reliably predicted geohazards like floods, and discovered previously unrecognized physical relationships.
Presenter: Chaopeng Shen | Civil & Environmental Engineering