Han Pu - Rice University Inspired by their great success in various areas, artificial neural networks have been introduced to study quantum many-body systems. Most of these works have employed shallow networks such as Restricted Boltzmann Machine, despite of the speculation that deep neural networks (DNNs) may possess many distinct advantages over shallow ones. The greatest challenge in applying DNN to quantum many-body calculation is to find an efficient algorithm to train the DNN which typically contains thousands of parameters. In our recent work, we constructed a flexible convolutional DNN and incorporated it into the variational Monte Carlo (VMC) method to solve a quantum spin-chain model. The key ingredient in our work is the development of an optimization algorithm that can efficiently train the network. The VMC with this optimization algorithm can be interpreted as a Reinforcement Learning process. Our work paves the way to study quantum many-body problems using DNNs.