Atmosphere Ocean Science Friday Seminar

---

Model training and generalization

Speaker: Adam Subel, CAOS

Location: Warren Weaver Hall 1314

Date: Friday, March 4, 2022, 4 p.m.

Notes:

To develop machine learning methods for SGS modeling that can find use in climate, we need models that can effectively generalize to conditions outside their training set. One particular requirement of this generalization is for networks to work across different resolutions or grids. In this talk, I will discuss applications of geometric deep learning to motivate a network design for such an application. Preliminary results using a graph neural network and message passing will be shown for the prediction of the subgrid SST forcing from CM2.6 data.