Breaking Scale Anchoring: Frequency Representation Learning for Accurate High-Resolution Inference from Low-Resolution Training
Published in preprint, 2025
We identify Scale Anchoring in zero-shot super-resolution spatiotemporal forecasting and propose Frequency Representation Learning to enable frequency-domain extrapolation from low-resolution training data.
Recommended citation: Wenshuo Wang, Fan Zhang. "Breaking Scale Anchoring: Frequency Representation Learning for Accurate High-Resolution Inference from Low-Resolution Training." preprint, 2025.