Presentation Profile
Artificial Intelligence–Driven Advances in Spectroscopic Analysis: From Signal Processing to Autonomous Interpretation
Currently Scheduled: 10/14/2026 - 1:00 PM - 2:00 PM
Room: Exhibit Hall Entrance
Main Author
Sayantani Sikder - Stony Brook University
- Raj Shah - Koehler Instrument Company, Inc.
Abstract:
Spectroscopic techniques including Raman, Fourier-transform infrared (FTIR), near-infrared (NIR), UV-visible, and X-ray photoelectron spectroscopy (XPS) remain essential tools for chemical and materials characterization, yet their analytical throughput is often limited by
manual spectral interpretation, batch variability, and the complexity of high-dimensional datasets. Recent advances in machine learning (ML) and deep learning (DL) have begun to overcome these limitations by enabling automated denoising, baseline correction, spectral deconvolution, and quantitative prediction with improved accuracy compared with conventional chemometric approaches. Convolutional neural networks and transformer-based architectures have demonstrated strong performance in Raman spectral classification, biomarker identification, and materials analysis, while physics-informed neural networks have improved both spectrum prediction and inverse structure determination. Semi-supervised and self-supervised learning strategies have further reduced reliance on large labeled datasets for UV-Vis and NIR spectroscopy, expanding opportunities for process analytical technology. Despite these advances, challenges remain in model interpretability, standardized benchmarking across spectroscopic modalities, and the limited availability of high-quality datasets, particularly for techniques such as XPS. This review summarizes recent developments in AI-enabled spectroscopic analysis, evaluates their advantages and limitations relative to traditional analytical workflows, and highlights emerging directions including physics-constrained learning and cross-modal foundation models for robust, interpretable spectroscopy.
