Abstract
One of the biggest challenges in harnessing AI methods in materials science is that the information landscape to discover multiscale connections is sparse and heterogeneous, and existing AI approaches have limited capacity to work in such data environments. Apart from hindering the development of effective predictive models, sparseness of data can also make the discovery of complex relationships that connect structure to the data-scarce property diffcult. Thus, addressing such sparse data problems requires us to utilize other sources of data that are relatively more abundant but presumably useful for the task at hand. In this article, we explore one strategy to address this issue based on the concept of a mixture of experts (MoE). We show that the MoE models can be used to predict a variety of mechanical properties. It is also shown that the MoE model can be used to uncover details linking the influence of site chemistry in perovskites on elastic constants.
Please click on the following link to read the full article:
https://link.springer.com/article/10.1557/s43577-024-00805-7