Fig. 4: Distinguishing strain and nematicity.
From: Machine learning the microscopic form of nematic order in twisted double-bilayer graphene
Predicted versus true values for the strain intensity ϵ (a) and angle θϵ (b). The prediction for the nematic intensities is depicted in panels c and d. The white dashed lines serve to guide the eye. R-squared (R2) and mean absolute percentage error (MAPE) metrics are shown in the inset. Details on how these metrics are calculated can be seen in the Methods section. The CNN architecture used to produce these results is described in Fig. 3a. Similarly to the prediction of the α parameter in the presence of only nematicity, outliers in θϵ are related to small ϵ.
