SupernoVAE is a non-linear method for dimensionality reduction. Dimensionality reduction methods are used to analyze and understand the underlying dynamics that drive the Earth System. Traditionally, Principal Component Analysis (PCA), also known as Empirical Orthogonal Functions, has been used for this purpose, but it fails when the relationship of features is non-linear. SupernoVAE aims to solve this by representing the data in a lower dimensional feature space where those relations are linear and consequently PCA works. The kernel function and its inverse are learned through a variational autoencoder. The method scales well and is also able to find linear relations.
Xavier Andoni Tibau Alberdi
Christian Requena-Mesa (FSU Jena- MPI Biogeochemistry)
Christian Reimers (FSU Jena)
Joachim Denzler (FSU Jena)
Markus Reichstein (MPI Biogeochemistry)
Veronika Eyring (DLR IPA, Uni Bremen)