UchUchU
cs.LGphysics.space-ph

Self-Supervised Representation Learning: From Spectral Foundation Models to Auroral Emission Spectra

Matthieu Le Lain, Gaël Cessateur, Sébastien Lefèvre

Abstract

Auroral spectrographs such as the Auroral Spectrograph In Skibotn (ASIS) record hundreds of thousands of emission spectra, but only a few hundred can be labelled by an expert. To exploit the rest, we pretrain a 1D Vision Transformer with a masked autoencoder on 223,000 unlabelled spectra. Without labels, its representation recovers the emission-line intensity ratios that physicists use to diagnose the precipitating particles (R^2 0.91 vs. 0.77 for an untrained control) and, under one linear prob…

Source：arXiv ／ PDF

Related news

Probabilistic Machine Learning for Short-Term Atmospheric Seeing Forecasting cs.LG · 2026.10.21 A Radiation-Hardened, Fault-Aware Quantized Neural Network That Self-Heals astro-ph.IM · 2026.10.20 Solar System Surface Technosignatures: Searching for Signs of Artificial Design on Extraterrestrial Terrain astro-ph.IM · 2026.10.20 Ammonia-Hydrogen Phase Equilibria in Planetary Envelopes from First-Principles-Based Machine-Learning Molecular Dynamics astro-ph.EP · 2026.10.19 Graph Neural Networks for Astrophysics and Cosmology astro-ph.IM · 2026.10.19 Native Geometry vs. Linear Recoverability: How Alignment Changes Platonic Scaling cs.LG · 2026.10.15

Looking for companies in this field?

UchUchU maintains a database of Japanese companies in the space industry.

Browse the database Read the entry guides
← Back to research trends