Fig. 1
From: A generative modeling approach for benchmarking and training shallow quantum circuits
General framework for data-driven quantum circuit learning (DDQCL). Data vectors are interpreted as representative samples from an unknown probability distribution and the task is to model such distribution. The 2N amplitudes of the wave function resulting from an N-qubit quantum circuit are used to capture the correlations observed in the data. Training of the quantum circuit is achieved by successive updates of the parameters θ, corresponding to specifications of single qubit operations and entangling gates. In this work, we use arbitrary single qubit rotations for the odd layers, and Mølmer-Sørensen XX gates for the even layers. At each iteration, measurements from the quantum circuit are collected and contrasted with the data through evaluation of a cost function which tracks the learning progress
