Fig. 2: The workflow of active learning loop.
From: Active learning for the power factor prediction in diamond-like thermoelectric materials
The loop contains three ingredients, Search Space including DFT database, machine learning module and DFT verification. Machine learning model is built based on the DFT database to recommend candidates, and then the candidates are verified with DFT calculations. If the extrapolation accuracy does not meet the convergence criterion, the verification data will be merged into the DFT database, and the whole loop will proceed.
