Faithfully loading classical data into a quantum system is a core problem in quantum machine learning and various quantum information processing tasks. In this work, we propose an efficient quantum autoencoder architecture that can construct a quantum state approximating the unknown classical distribution with high precision and with only linear circuit depth. Simulation experiments show that our proposed method substantially outperforms state-of-the-art methods on a wide range of datasets by evaluating divergences between the loaded distributions and the target distribution, and it also enjoys a faster convergence rate and stability. Moreover, the proposed scheme can be efficiently implemented on near-term hybrid classical-quantum systems with very shallow circuit depths.

Score function-based natural language generation (NLG) approaches such as REINFORCE, in general, suffer from low sample efficiency and training instability problems. This is mainly due to the non-differentiable nature of the discrete space sampling and thus these methods have to treat the discriminator as a black box and ignore the gradient information. To improve the sample efficiency and reduce the variance of REINFORCE, we propose a novel approach, TaylorGAN, which augments the gradient estimation by off-policy update and the first-order Taylor expansion. This approach enables us to train NLG models from scratch with smaller batch size --- without maximum likelihood pre-training, and outperforms existing GAN-based methods on multiple metrics of quality and diversity.