Fig. 8
A simplified workflow for new AI-based material design. Firstly, the necessary data for modeling and model evaluation is collected, including experimental data, simulation data, and existing literature data. Then, the collected data is preprocessed through the steps of data cleaning and normalization. Next, the data set is divided into training sets, verification sets, and test sets using k-fold cross-entropy validation. And the indexes of new material performance evaluation are summarized into a confusion matrix by one-hot encoding. Secondly, the appropriate model algorithm is selected for training according to the research problem, and HPC is used if necessary. Evaluate the performance of the model using established criteria and iteratively adjust the model parameters and use techniques to optimize performance under the guidance of these standards. Finally, the forecast results are discussed and analyzed reasonably
