Validation and interpretability of data-driven models

This is a Whitepaper submitted to the 2017 DOE ASCR Applied Math Meeting. It addresses research topics in the "Convergence of data- and model-driven discovery" subject area. In particular, it proposes research activities that would enhance the interpretability of data-driven models, such as neural nets, which are increasing being used in multiscale simulations for upscaling/downscaling operations e.g., as turbulence closures etc. The research would allow us validate such empirical, data-driven models against physics theory