Data and Code for: Chemodynamical Outlier Detection in Gaia DR3 Using β-Variational Autoencoders

This dataset accompanies the Research Note "Chemodynamical Outlier Detection in Gaia DR3 Using β-Variational Autoencoders" (Robles Leyton 2026, RNAAS). Contents: - Top 21 chemodynamical outliers identified by β-VAE reconstruction error - β-VAE training code (PyTorch implementation) - Paper manuscript (LaTeX) and diagnostic figure - Full documentation and requirements Key results: - 70.7% of classical outliers recovered - 29.3% additional candidates flagged (missed by traditional cuts) - 2 stars with disk chemistry + halo dynamics (heated disk candidates) The method learns correlations between 9 orbital features (e, z_max, L_z, E, U, V, W, R, z) and identifies rare combinations through unsupervised anomaly detection. Data sources: Gaia DR3, LAMOST DR7, APOGEE DR17.