This talk will introduce theory-guided data science, a novel paradigm of scientific discovery that leverages the unique ability of data science methods to automatically extract patterns and models from data, but without ignoring the treasure of knowledge accumulated in scientific theories. Theory-guided data science aims to fully capitalize the power of machine learning and data mining methods in scientific disciplines by deeply coupling them with models based on scientific theories. This talk will describe several ways in which scientific knowledge can be combined with data science methods in various scientific disciplines such as hydrology, climate science, aerospace, and chemistry. To demonstrate the value in combining physics with data science, the talk will also introduce a novel framework for combining deep learning methods with physics-based models, termed as physics-guided neural networks, and present some preliminary results of this framework for an application in lake temperature modeling. The talk will conclude with a discussion of future prospects in exploiting latest advances in deep learning for building the next generation of scientific models for dynamical systems, where theory-based and data science methods are used at an equal footing.
Back to Workshop III: HPC for Computationally and Data-Intensive Problems