Zoom (Zugangsdaten unter Bemerkungen)

We present a machine learning approach for developing stochastic parametrizations of subgrid fluxes in finite-volume discretizations of PDEs expressed in flux form. In particular, we utilize Generative Adversarial Networks (GANs) to parametrize subgrid fluxes in equations for coarse variables defined via spatial filtering. We discuss how to construct and train the network and demonstrate that our approach reproduces statistical properties of coarse variables in fully resolved simulations. Moreover, we also demonstrate that GANs are robust and can be utilized without re-training in new regimes (e.g. changes in forcing). We use the Burgers equation to illustrate our approach.

https://tu-dortmund.zoom.us/j/97448817932?pwd=cUdoZi9UMEcrUWFlS3NsSzBpMnIxdz09 Meeting ID: 974 4881 7932 Passcode: 181775

Prof. Dr. Ilya Timofeyev

Houston