Fig. 4: Effect of training set size on various metrics.
From: A neural network-based synthetic diagnostic of laser-accelerated proton energy spectra
a Synthetic diagnostic ensemble proton spectra prediction PUV (green), (b) ΦP prediction PUV (green), (c) sharpness of predicted spectrum (green) and ΦP (blue) distributions, and (d) calibration error (CE) and signed CE, all as a function of training set size. Error bars represent variability from repeated resampling and retraining. Plots (a, b) include power-law curve fits extrapolated to a training set size of 3600, and include the performance and power-law fit of the surrogate model (red).
