The 225th R-CCS Cafe - part 3

Bayesian approach, which allows optimal data-augmentation by backward prediction, has become a popular approach for accelerating material design. Although, Bayesian optimization usually requires experimental sampling, I have employed data-driven surrogate model for forward prediction to augment the database with virtual sampling.
Surrogate model construction has been aided by quantum chemistry calculations to improve the prediction.
One application is hole transporting material for perovskite solar cell. Many experiments have been conducted on HTMs that shares the characteristics of spiro-OMeTAD, the most recognized HTM, to improve energy conversion efficiency and reduce fabrication costs of solar cells. Extensive amount of experimental data is already available for database construction. In this talk, I will present how the data-driven surrogate model and the generated virtual samples successfully captures the features of known HTMs, along with application to high-performance biopolymer design.