SPACIOUS is expanding its scientific toolbox with the release of GANDALF, GUASOM, and GAVS+ — three new services designed to support state-of-the-art machine learning, spectral analysis, and large-scale data exploration. These tools strengthen SPACIOUS as a versatile environment for working with complex astronomical datasets, including those from Gaia.
GANDALF: Adversarial Disentangling for Stellar Spectra
GANDALF is an open, python framework built on an adversarially trained autoencoder that disentangles key physical parameters in stellar spectra. By assigning an independent discriminator to each parameter. such as temperature, gravity, or chemical composition, the method avoids the limitations of standard Generative Adversarial Network (GAN) approaches, achieving cleaner latent-space separation and substantial dimensionality reduction. Tests on synthetic and Gaia data show accurate parameter recovery and robust reconstruction performance.
GUASOM: Self-Organizing Map Analysis for Large Spectroscopic Data
GUASOM (Gaia Utility for the Analysis of Self-Organizing Maps) provides an intuitive interface for training and visualising SOMs on massive spectroscopic or spectrophotometric datasets. It offers an effective way to explore structure, identify patterns, and detect anomalies across high-dimensional data collections.
GAVS+: Next-Generation Gaia Visualisation
GAVS+ builds on the Gaia Archive Visualisation Service to deliver a cloud-ready, high-performance environment for interactive exploration of multi-billion-row tables. New features include support for user-uploaded tables, arbitrary attribute combinations, and full Jupyter notebook integration, enabling flexible and reproducible visual analytics workflows.
