Quantum machine learning of graph-structured data

Graph structures are ubiquitous throughout the natural sciences. Here we develop an approach that exploits quantum source's graph structure to improve learning via arbitrary neural network (QNN) ansatz. In particular, devise and optimize a self-supervised objective capture information-theoretic closeness of states in training QNN. Numerical simulations show our improves efficiency generalization behavior base On practical note, scalable implementations procedure described this paper likely feasible on next generation computing devices.