Recent strides in artificial intelligence for predicting thermoelectric properties and materials discovery

Abstract

Machine learning models as part of artificial intelligence have enjoyed a recent surge in answering a long-standing challenge in thermoelectric materials research. That challenge is to produce stable, and highly efficient, thermoelectric materials for their application in thermoelectric devices for commercial use. The enhancements in these models offer the potential to identify the best solutions for these challenges and accelerate thermoelectric research through the reduction in experimental and computational costs. This perspective underscores and examines recent advancements and approaches from the materials community in artificial intelligence to address the challenges in the thermoelectric area. Besides, it explores the possibility for these advancements to surpass existing limitations. Additionally, it presents insights into the material features influencing model decisions for thermoelectric property predictions and in some cases new thermoelectric material discovery. In the end, the perspective addresses current challenges and future potential studies beyond classical ML studies for thermoelectric research.

Keywords

• algorithms
• machine learning
• thermoelectric
• XGBoost
• materials informatics