Leveraging Transfer Learning and Chemical Principles toward Interpretable Materials Properties

Machine learning is emerging as a new paradigm to rationalize chemical properties for deepening our understanding of chemistry and providing instructive clues on better materials performance. While the complex architecture of machine learning contributes to unprecedented capability in this task, it prevents easy interpretation, leading to extensive criticisms on the lack of physical foundations for the black-box like models. Here, we demonstrate a transfer learning strategy that leverages fundamental principles of chemistry to offer adequate physical insights for the interpretation. Through interpreting the models for the formation energies of inorganic compounds, the proposed strategy revealed the deficiency of deep neural network in handling interelemental patterns and proved the more proper abstraction of recurrent neural network with attention mechanism, which led to predicting the elegant form of periodic table with high precision. The success demonstrates a new solution toward models with full transparency in materials informatics.

Automated summary

This study demonstrates a transfer learning strategy that utilizes fundamental principles of chemistry to offer physical insights for interpreting machine learning models. It reveals the limitations of deep neural networks in handling interelemental patterns and highlights the suitability of recurrent neural networks with attention mechanism for predicting the periodic table with high precision. This success presents a new approach towards models with full transparency in materials informatics.