Abstract |
Benefited from recent advances in big-data analytics, the machine learning method was proposed to accelerate discovery of materials with desired properties. In this talk, we apply several data-driven algorithms/strategies to establish high-throughput models that allows ready and accurate prediction on the Seebeck coefficient, the lattice thermal conductivity, and the ZT values of thermoelectric materials. Without any input from first-principles calculations, the models only require the information of crystal structures or fundamental properties of the constituent atoms, and can be readily generalized to systems drastically beyond the training data. Our work not only provides a large space for exploring high-performance thermoelectric materials, but also attests to the increasing importance of artificial intelligence-based approaches in modern materials discovery.
|
