Variational optimization in the AI era: supervised wave-function optimization and computational graph states.

An important approach to the quantum many-body problem is to write down a compact variational ansatz which represents the target quantum state. The success of a particular model depends on its ability to capture the structure of the state and optimize all variational parameters. We introduce a machine learning inspired computational framework that works with wave-functions represented as differentiable computational graphs and develop a novel optimization algorithm Supervised Wave-function Optimization, that allows for effective optimization of such models. We present results on several architectures showing the efficiency of our approach.