AI and the World's Hard Problems · Astronomy & space · arXiv · 2026-10-26
OmniSpectra: A Unified Foundation Model for Universal Spectra Representation Learning
Standard sequence models index positions by token rank, which forces spectra from different instruments to be resampled and distorts their physical coordinates. The authors propose OmniSpectra, a wavelength-native masked autoencoder pretrained on eight surveys. With only 12.2M parameters, it reaches state-of-the-art results against seven baselines and outperforms substantially larger models.