Bachelor Thesis: Solving quantum many-body system based on transfer reinforcement learning

Abstract

A single convolutional neural network (CNN) can be applied to systems with different sizes. Based on this property, we proposed a transfer learning scheme for the neural network quantum state in which the CNN was frstly trained by the exact diagonalization result in a small system, and then applied to the larger system for further variational optimization. This transfer learning scheme improved the accuracy and the stability in the 1D frustrated J1-J2 model compared with direct variational optimization.