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

2026-05-23 · Source: cs.LG updates on arXiv.org · Field: Science & Research — Physical Sciences & Chemistry, Mathematics & Computational Sciences, Research Methodology & Innovation · Depth: Expert, extended

Summary

Near-infrared spectroscopy (NIRS) is a rapid, non-destructive chemical sensing technology, but its practical deployment requires robust calibration models for high-dimensional, collinear spectra, limited sample sizes, and outliers. This study evaluates tabular foundation models, specifically TabPFN-2.5, as a new calibration strategy on 66 NIR datasets (54 regression, 12 classification). Using a unified validation framework, preprocessing-optimized TabPFN (TabPFN-opt) achieved the best overall average rank (1.800) in regression, significantly outperforming PLS, CatBoost, raw-spectra TabPFN, and CNN-1D, while being statistically comparable to Ridge regression. In classification, TabPFN applied directly to raw spectra (TabPFN-Raw) provided the best average rank (2.000). Robustness analyses indicated TabPFN's advantage diminished on spectral outliers and extrapolated samples. The findings suggest tabular foundation models can complement established chemometric workflows for small- to medium-sized NIRS datasets, emphasizing the need for spectroscopy-specific priors.

Key takeaway

For Machine Learning Engineers developing NIR chemical sensing solutions, you should integrate tabular foundation models like TabPFN into your workflows. Optimize preprocessing, particularly with Savitzky-Golay transformations for regression tasks, to achieve superior average predictive performance. Be aware that classical chemometric models may still offer better robustness for spectral outliers or extrapolated samples, so consider hybrid approaches for critical applications.

Key insights

Tabular foundation models like TabPFN can enhance NIR chemical sensing calibration, especially with optimized preprocessing for regression tasks.

Principles

• Preprocessing significantly impacts TabPFN's regression performance.
• Robustness to outliers differs from average predictive performance.
• TabPFN offers competitive performance without dataset-specific training.

Method

The study used a two-stage hierarchical preprocessing search strategy and a three-fold cross-validation for joint preprocessing and hyperparameter optimization on calibration data, followed by external test evaluation.

In practice

• Integrate TabPFN with Savitzky-Golay smoothing for NIRS regression.
• Consider classical chemometric models for outlier-prone or extrapolated samples.
• Use TabPFN for small-to-medium NIRS datasets to reduce training time.

Topics

• Near-Infrared Spectroscopy
• Tabular Foundation Models
• TabPFN
• Chemometrics
• Preprocessing Optimization
• Regression Analysis
• Classification Tasks

Code references

• GBeurier/nirs4all

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• A Causal Foundation Model for Structure and Outcome Prediction — TabPFN-CFM is a causal foundation model predicting structure and outcomes, supporting Pearl's Causal Hierarchy queries.
• Reframing preprocessing selection as model-internal calibration in near-infrared spectroscopy: A large-scale benchmark of operator-adaptive PLS and Ridge models — Embedding preprocessing selection within NIRS calibration models enhances robustness, interpretability, and efficiency.
• RxnNano:Training Compact LLMs for Chemical Reaction and Retrosynthesis Prediction via Hierarchical Curriculum Learning — Deep chemical understanding and structured training enable compact LLMs to outperform larger models in reaction prediction.
• Introducing TabFM: A zero-shot foundation model for tabular data — TabFM applies zero-shot in-context learning to tabular data, eliminating manual tuning and feature engineering for classification and regression.
• When Tabular Foundation Models Meet Strategic Tabular Data: A Prior Alignment Approach — Strategic manipulation creates a prior mismatch for tabular foundation models, which SPN corrects at inference.
• GOTabPFN: From Feature Ordering to Compact Tokenization for Tabular Foundation Models on High-Dimensional Data — GOTabPFN enhances small tabular foundation models for High-Dimensional, Low-Sample Size data via graph-guided feature ordering and neuro-inspired subunit compression.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.