1: \begin{abstract}

2: Inspired by the success of generative adversarial networks (GANs) %, a 

3: in learning high-dimensional distributions, % quantum mechanical counterparts of GANs, 

4: recently quantum GAN (QuGAN) has been  proposed as quantum algorithm to learn quantum states. Whether QuGAN is capable of performing meaningful tasks in realistic physical systems, however, is less understood. In this work, we propose a new variant of QuGAN, \text{Hamiltonian} QuGAN (HQuGAN), that generates desired quantum resources by controlling time-continuous quantum dynamics using quantum optimal control. Such continuous pulse model not only can fit naturally to near-term hardwares, but also has the potential to provide a better convergence of the algorithm due to  overparameterization comparing to the gate-based model. Numerically we show that HQuGAN successfully learns various quantum states up to nine  qubits. % including $9$-qubits GHZ state. 

5: Furthermore, we describe how to implement HQuGAN on quantum computers and analyze its computational cost. We also show how QuGAN algorithm can be modified to learn unitary evaluations. 

6: 

7: \end{abstract}

8: