Abstract
Ian Langmore (Google): From scoring rules to probabilistic ML weather forecasting, an interactive tutorial
This workshop starts by considering probabilistic loss functions as an alternative to likelihood-based modeling. After a theoretical treatment of proper scoring rules, we move on to ML-enhanced weather forecasting as an application. The discussion is centered on the NeuralGCM model, which uses a differentiable fluid solver in conjunction with a neural network. Finally, we demonstrate weather forecasts in a notebook using open source NeuralGCM and JAX.
Back to the Agenda