Automatically generating RANS models from DNS data
The goal here is to generate a PDE that effectively closes the RANS equations using DNS data. We will generate a bunch of PDE terms and solve for the coefficients that make the equation balance at many points in space (at least as many as the number of additional terms). We can then drop terms with very small coefficients.
If there is a transparent way for saving ML models and making predictions on other platforms, we could remove the requirement that the model be a PDE. I believe there is a bunch of work being done on this at the moment.
