A new interpretable AI approach shows how models connect atomic structure to optical spectra.
Japanese researchers have developed an interpretable AI method that explains how AI models make predictions in materials discovery. This method analyzes features learned by a trained AI model and uses them to identify relationships between atomic structures and optical spectra.
The study was led by researchers at Tokyo University of Science in collaboration with Tohoku University. The research will be published in the journal Advanced Intelligent Discovery.
AI is increasingly used in materials research to predict how materials will behave based on their atomic structure. Such models can accelerate materials discovery and reduce reliance on trial-and-error experiments, but many operate as black boxes, making it difficult to understand how specific predictions are arrived at.
The researchers addressed this problem by analyzing a trained AI model that predicts optical absorption spectra from atomic structure data. They extracted features from the model’s internal layers and clustered materials according to common structural and spectral properties.
The research team used an atomistic line graph neural network trained on data from 2,681 metal oxides, chalcogenides, and related compounds. The clustering process classified materials into groups that share structural properties such as elemental composition, atomic coordination, bond lengths, bond angles, and similar spectral signatures.
According to the researchers, the model learned meaningful relationships between atomic structure and material properties without explicitly providing oxidation state or electronic configuration as input. Therefore, interpretable AI methods can help researchers identify the factors behind desired spectral shapes and support more rational material design.
This approach can also be applied beyond the optical absorption spectrum. The researchers said their approach could also help explain how atomic arrangement affects other material properties under different conditions such as temperature and pressure, opening new possibilities for designing materials with targeted properties.
Why is it important?
One of the main challenges facing the use of AI in scientific research is explainability. While AI systems can identify patterns and generate accurate predictions, researchers often need to understand the reasoning behind their predictions before they can confidently apply them to experimental settings.
By revealing how AI models connect atomic structure and material properties, interpretable AI has the potential to make machine learning a more effective tool for scientific discovery. This approach could help accelerate the development of advanced materials for a variety of applications, from renewable energy and electronics to sensors and next-generation manufacturing, while increasing confidence in AI-assisted research.
Want to learn more about AI, technology, and digital diplomacy? Then contact Diplo chatbot.
