Image Details
Caption: Figure 4.
Per-r-magnitude-bin point performance on the DR6 test set at ntrain = 121,626 for the TabPFN 2.5 and the two other best-performing methods. Top panel: per-bin RMSE. Bottom panel: per-bin outlier fraction η0.15. Shaded bands are ±1 interrepetition standard deviation across the five seeds. TabPFN 2.5 is strongest at bright magnitudes and retains a modest advantage in the faint tail; GBM-Point is generally worse, particularly for the outlier fraction.
© 2026. The Author(s). Published by the American Astronomical Society.