RamanGPT

RamanGPT applies machine learning to Raman spectroscopy of two-dimensional materials such as transition-metal dichalcogenides (e.g., VSe₂, WS₂). It establishes a bidirectional mapping between crystal structures and Raman spectra: a forward model (graph neural networks) that predicts spectra from atomic structures, and an inverse model (generative transformers) that infers structure from measured spectra. Trained on a mix of experimental and simulated spectra, the goal is to make Raman a faster, more quantitative probe of 2D-material structure and quality.

This work is described in RamanGPT: Bidirectional Mapping Between Crystal Structures and Raman Spectra with Graph Neural Networks and Generative Transformers (arXiv:2606.03764).