Explainable Artificial Intelligence Approach to Identify the Origin of Phonon‐Assisted Emission in WSe2 Monolayer

The application of explainable artificial intelligence in nanomaterial research has emerged in the past few years, which has facilitated the discovery of novel physical findings. However, a fundamental question arises concerning the physical insights presented by deep neural networks; the model interpretation results have not been carefully evaluated. Herein, explainable artificial intelligence and quantum mechanical calculations is bridged to investigate the correlation between light scattering and emission in a WSe2 monolayer. Convolutional neural networks using light scattering and emission data are first trained, while expecting the networks to determine the relationships between them. The trained models are interpreted and the specific phonon contribution during the exciton relaxation process is derived. Finally, the findings are independently evaluated through quantum mechanical calculations, such as the Born–Oppenheimer molecular dynamics simulation and density functional perturbation theory. The study provides reliable fundamental physical insight by evaluating the results of neural networks and suggests a novel methodology that can be applied in materials science. A combination of explainable artificial intelligence‐applied correlative spectroscopy and quantum mechanical calculations enables the discovery and validation of novel physics. Deep neural networks using Raman scattering and photoluminescence images are trained, and their interpretation results provide novel physics. Molecular dynamics and density functional perturbation theory calculations are calculated to validate the physical findings.