This paper introduces a new neural network, GQNQ, capable of learning multiple quantum states from classically simulated data. Unlike existing networks trained on experimental data from the specific quantum state to be characterized, GQNQ can be trained offline on a fiducial set of states and measurements and then used to characterize other states with structural similarities. The network builds its own data-driven representation of a quantum state and uses it to predict outcome statistics of unperformed measurements. The generated state representations are also applicable to quantum state clustering and the identification of different phases of matter.

Yan Zhu, Ya-Dong Wu, Ge Bai, Dong-Sheng Wang, Yuexuan Wang, Giulio Chiribella