Comparison of the reconstruction accuracy of the nonlinear CNN with a linear model on the test dataset. The solid line indicates the mean value, and the colored region shows the 68% and 95% confidence intervals. The dotted line indicates where the RMS error of the reconstruction is equal to the input RMS, i.e., where there is no improvement. Left: residual RMS error of the prediction as a function of the RMS of the input wavefront, shown on a linear scale. Right: improvement factor, defined as the ratio between the residual RMS and the input RMS, as a function of the input RMS, which is shown on a logarithmic scale.