Generative adversarial networks constitute a powerful approach to generative modeling. While generated samples often are indistinguishable from real data, there is no guarantee that they will follow the true data distribution. In this work, we propose a method to ensure that the distributions of certain generated data statistics coincide with the respective distributions of the real data. In order to achieve this, we add a Kullback-Leibler term to the generator loss function: the KL divergence is taken between the true distributions as represented by a conditional energy-based model, and the corresponding generated distributions obtained from minibatch values at each iteration. We evaluate the method on a synthetic dataset and two real-world datasets and demonstrate improved performance of our method.

翻译：生成对抗网络构成了生成建模的一种强大方法。尽管生成的样本通常与真实数据难以区分，但无法保证它们会遵循真实的数据分布。在这项工作中，我们提出了一种方法来确保某些生成数据统计量的分布与真实数据的相应分布一致。为了实现这一目标，我们在生成器损失函数中添加了一个Kullback-Leibler项：该KL散度是在由条件能量模型所表示的真实分布与每次迭代中从小批量值获得的相应生成分布之间计算的。我们在一个合成数据集和两个真实世界数据集上对该方法进行了评估，并证明了我们方法的性能提升。