Exploration of groups and outliers in Gaia RVS stellar spectra with metric learning

The Gaia mission is transforming our view of the Milky Way by providing distances towards a billion stars, and much more. The third data release includes nearly a million spectra from its Radial Velocity Spectrometer (RVS). Identifying unexpected features in such vast datasets presents a significant challenge. It is impossible to visually inspect all of the spectra and difficult to analyze them in a comprehensive way. In order to supplement traditional analysis approaches, and in order to facilitate deeper insights from these spectra, we present a new dataset together with an interactive portal that applies established self-supervised metric learning techniques, dimensionality reduction, and anomaly detection, to allow researchers to visualize, analyze, and interact with the Gaia RVS spectra in straightforward but under-utilized manner. We process 312,295 Gaia DR3 RVS spectra (from 997,162 available spectra) by focusing on the high-quality subset with SNR > 50, which reduces noise-dominated artifacts in the anomaly ranking and makes the identified outliers more interpretable. We demonstrate a few example interactions with the dataset, examining groupings and the most unusual RVS spectra, according to our metric. This combination of methodology and public availability enables broader exploration, and may reveal yet-to-be-discovered stellar phenomena.