Quantum Generative Adversarial Networks

Quantum machine learning could be one of the first applications for general-purpose quantum computers in the near future. Given the computational complexity of the quantum spectrum problem, quantum circuits outperform conventional neural networks in terms of expression. The notion of generative adversarial training, in order to form a distinct generative model, is a major recent achievement in classical machine learning. This paper describes a method to implement generative adversarial networks on quantum computer to minimize the loss/error of generator model. The paper implements hybrid-quantum architecture for GAN using the minimax game strategy for loss calculation. As a generator, it uses a quantum circuit, and as a discriminator, it uses a classical neural network. Only one-qubit rotation gates and controlled two-qubit phase gates are used in the parametric quantum circuit in the proposed approach.