Tabular foundation models for robust calibration of near-infrared chemical sensing data

In three linesBenchmark of TabPFN (tabular foundation model) on 66 NIR datasets (54 regression, 12 classification tasks). Optimized TabPFN outperforms PLS, CatBoost, and CNN-1D in regression; matches Ridge in classification. Advantage diminishes on spectral outliers and extrapolation.

Summary generated by Claude — human-verified