Predictive descriptors in machine learning and data-enabled explorations of high-entropy alloys

Located at the intersection of intriguing material phases and potentially superior mechanical properties, highentropy alloys (HEAs) have been gaining increasing interest across academia and industry, in particular for high temperature applications. The extremely vast compositional space (similar to 10(12) possibilities) for these complex metallic alloys require rigorous predictive strategies to scavenge the expansive realm of unexplored alloy composition-processing-structure-property landscape. Enabled by the advances in artificial intelligence and machine learning methods, data-driven exploration of HEAs are burgeoning, not only for the discovery of new materials but also for predicting properties that are challenging to measure using experiments or require resource and time-intensive computations. Nevertheless, success of such data-enabled models in delivering accurate estimates of microstructures and properties depend on the choice of appropriate descriptors that suitably represent the underlying structural and transport mechanisms. This review provides a synopsis of the contemporary and effective data-centric methods employed to examine HEAs, with special emphasis on the selection and role of feature descriptors. We highlight some of the current challenges with these approaches that the computational materials community is facing, and offer recommendations to address them.