Figure 5
Final noise level \(\lambda\) as a function of the inverse temperature \(\beta\) for various models and system sizes. We used a symlog scale for the y-axis, hence the scale becomes linear below \(10^{-3}\). We observe a clear decrease of the noise level with \(\beta\), with \(\lambda \approx 1\) for \(\beta =10^{-3}\) (corresponding to the maximally-mixed state) and \(\lambda \approx 0\) for \(\beta \approx 10^2\) (corresponding to the ground-state). It shows that the general relationship between the noise and the temperature has overall been correctly learned by our model.
