Fig. 10: Training dynamics of total error ϵα(t) on IBM quantum devices, Kyiv.
From: Quantum-data-driven dynamical transition in quantum learning
In (a, b), the target values are chosen to be y1 = − 0.3, y2 = − 3 and y1 = − 1, y2 = − 3 separately, corresponding to the mixed-frozen dynamics and critical-frozen-error dynamics. Solid light blue and purple curves represent experimental results for ϵ1(t) and ϵ2(t), dashed dark blue and pink curves represent corresponding ideal simulation results. An n = 2 qubit D = 6-layer hardware efficient ansatz (with L = 24 parameters) is utilized to minimize loss function with input states \(\left\vert {\psi }_{1}\right\rangle =\left\vert 01\right\rangle\), \(\left\vert {\psi }_{2}\right\rangle =\left\vert 10\right\rangle\), and the observable is \(\hat{O}={\hat{\sigma }}_{1}^{z}\), Pauli-Z operator on the first qubit.
