Training#
Mean Squared Error#
The Mean Squared Error (MSE) over ten parameter initialisation seeds and ten randomly generated problem instances during training. Lines represent the mean, and shaded areas show the standard deviation over the 10 × 10 configurations.
Coefficient Difference#
The difference \(\Delta_{c_{\boldsymbol{\omega}}}\) between the target and learned coefficients \(c_{\boldsymbol{\omega}}(\boldsymbol{\theta})\) and \(c'_{\boldsymbol{\omega}}\) respectively of the QFMs averaged over ten model parameter initialisation seeds and ten randomly generated problem instances during training. Lines represent the mean, and shaded areas show the standard deviation over the 10 × 10 configurations.
Coefficients during Training for Problem Instances#
In the following, we show the learned absolute coefficients of the QFMs averaged over ten parameter model initialisation seeds for each single randomly generated probelm instance with seeds \([1000\dots 1009]\) and different types of noise (3%) during training of 1000 steps. In the figures the shaded areas correspond to the standard deviation across the parameter initialisation seeds from the mean. The coefficients of the objective Fourier series are marked with dashed lines.
Problem instance seed 1000 (Our Figure 18 in the paper):

Entanglement during Training#
Entangling capability, assessed with the Meyer-Wallach measure for pure states (noiseless and coherent gate errors) and entanglement of formation for mixed states (decoherent-, SPAM- and damping errors) during training. Lines represent the mean over ten parameter initialisation seeds and ten problem generation seeds. Shaded areas represent the standard deviation.























