Simulation-to-real transfer learning for infrared spectroscopic chemical sensing and analysis from molecules to complex samples

TL;DR - UltraIR is a 100M+ parameter foundation model pretrained on roughly 60 million simulated infrared spectra, then adapted to diverse chemical-analysis tasks. It enables data-efficient simulation-to-real transfer and cross-instrument zero-shot inference.

• Pretraining combines spectral reconstruction, molecular-fingerprint alignment, and functional-group prediction.
• Applications span molecular analysis, mixtures, bacteria, medicinal herbs, microplastics, and soil properties.
• UltraIR outperforms conventional machine-learning and task-specific deep-learning baselines across the reported tasks.
• It remains effective with limited labeled experimental data and transfers across spectrometers and laboratories.