GANterpolate
Generative reconstruction of sparse scientific data
GANterpolate is a hybrid U-Net GAN and interpolation framework for reconstructing missing values in sparse spatiotemporal data. On ERA5 climate data it reduced temperature RMSE from 6.84 to 5.37. GANterpolate+ generalizes the approach across agriculture, energy, healthcare and finance, outperforming baselines in all five domains.
Best Paper Award, IEEE UEMCON 2025. Extended version published at IEEE CCWC 2026.