Figure 2.

Process of a BuRNN simulation that includes adaptive sampling. At every xth time step during MM, two neural networks (A and B) are compared. When predictions diverge, the training set is expanded by additional QM calculations.