Verner Soh

High entropy alloys (HEAs) are highly sought after in the aerospace and energy storage sectors with their outstanding mechanical and physical properties. The discovery of HEAs through machine learning (ML) techniques has emerged as a transformative approach to navigate their vast compositional space. ML models like artificial neural networks have demonstrated superior predictive capabilities for mechanical and thermal properties compared to traditional methods (e.g. Density functional theory). These models operate on active learning strategies to uncover promising alloys. However, the optimization process faces inherent difficulties: an increasing number of principal elements warrants consideration of more features and processing of larger datasets, while small experimental datasets limit model generalizability. Additional challenges arise where ML models still rely heavily on the data training set and their progression in generating compositions beyond the provided dataset. Recent advances in high-throughput characterization and generative ML models show promise for accelerating discovery, yet persistent gaps in understanding phase stability rules continue to hinder fully independent alloy design. This review highlights the progress in ML-driven HEA discovery while examining the difficulties with current ML strategies.