Fig. 2.

The plots visualize the upper and lower bound (Proposition 1) of the correlation between the surrogate model $g_{\theta }$ and the physics model f when the surrogate model $g_{\theta }$ is trained on a related data source h that is correlated to f. The upper and lower bounds depend on the correlation between the data source h and the physics model f (“quality of the data”) as well as correlation between the data source h and the surrogate model $g_{\theta }$ (“quality of training”).