Accounts of Materials Research

While the uptake of data-driven approaches for materials science and\nchemistry is at an exciting, early stage, to realise the true potential of\nmachine learning models for successful scientific discovery, they must have\nqualities beyond purely predictive power. The predictions and inner workings of\nmodels should provide a certain degree of explainability by human experts,\npermitting the identification of potential model issues or limitations,\nbuilding trust on model predictions and unveiling unexpected correlations that\nmay lead to scientific insights. In this work, we summarize applications of\ninterpretability and explainability techniques for materials science and\nchemistry and discuss how these techniques can improve the outcome of\nscientific studies. We discuss various challenges for interpretable machine\nlearning in materials science and, more broadly, in scientific settings. In\nparticular, we emphasize the risks of inferring causation or reaching\ngeneralization by purely interpreting machine learning models and the need of\nuncertainty estimates for model explanations. Finally, we showcase a number of\nexciting developments in other fields that could benefit interpretability in\nmaterial science and chemistry problems.\n

Interpretable and Explainable Machine Learning for Materials Science and Chemistry