KAN4Met: Kolmogorov–Arnold Networks for weather forecasting
Published:
Hello dear readers,
A new project started this summer: KAN4Met. It’s funded by Spain’s Agencia Estatal de Investigación (PID2024-155187OB-I00) and led by David Gómez-Ullate. I’m part of the team.
The idea is to test Kolmogorov–Arnold Networks (KANs) for weather forecasting. A KAN is a neural network, but its learned functions sit on the connections, so you can plot them and read them. That’s the part I like. When a forecast feeds a decision, I want to see what the model learned, not only how accurate it is.
I’m interested in how they can feed into weather routing for ships, which is where most of my work happens.
It’s early days, so no results to share yet. I’ll post updates as we have them. There is also a project page.
All the best,
Daniel Precioso, PhD
