Thermal conductivity prediction from basic properties using a developed artificial neural network

High-throughput screening and material informatics have shown a great power in material discovery including Li-battery materials, alloys, photocatalysts, and nanowires. In this talk, we will present the accurate thermal conductivity prediction from machine learning technique using a developed artificial neural network (ANN). With 231 datasets of the basic properties describing materials calculated from first-principles and the corresponding thermal conductivity from experimental measurements as training data, the constructed ANN is well trained by iterating to reduce the loss function. The trained ANN model for thermal transport successfully captures the general correlation between basic properties and thermal conductivity for different types of materials, which is predictive spanning 4 orders of magnitude of the thermal conductivity. The developed ANN model in our work for fast and accurately predicting thermal conductivity provides a powerful tool for the large-scale thermal material screening with targeted thermal transport property.