We introduce a new algorithm named WGAN, an alternative to traditional GAN training. In this new model, we show that we can improve the stability of learning, get rid of problems like mode collapse, and provide meaningful learning curves useful for debugging and hyperparameter searches. Furthermore, we show that the corresponding optimization problem is sound, and provide extensive theoretical work highlighting the deep connections to different distances between distributions.

The goal of this paper is not to introduce a single algorithm or method, but to make theoretical steps towards fully understanding the training dynamics of generative adversarial networks. In order to substantiate our theoretical analysis, we perform targeted experiments to verify our assumptions, illustrate our claims, and quantify the phenomena. This paper is divided into three sections. The first section introduces the problem at hand. The second section is dedicated to studying and proving rigorously the problems including instability and saturation that arize when training generative adversarial networks. The third section examines a practical and theoretically grounded direction towards solving these problems, while introducing new tools to study them.

Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training instability. The recently proposed Wasserstein GAN (WGAN) makes progress toward stable training of GANs, but sometimes can still generate only poor samples or fail to converge. We find that these problems are often due to the use of weight clipping in WGAN to enforce a Lipschitz constraint on the critic, which can lead to undesired behavior. We propose an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input. Our proposed method performs better than standard WGAN and enables stable training of a wide variety of GAN architectures with almost no hyperparameter tuning, including 101-layer ResNets and language models with continuous generators. We also achieve high quality generations on CIFAR-10 and LSUN bedrooms. † * Now at Google Brain † Code for our models is available at https://github.com/igul222/improved_wgan_training.

我们介绍了用于生成建模的广义能量模型（GEBM）。这些模型组合了两个训练有素的组件：基本分布（通常是隐式模型），可以在高维空间中学习具有低固有尺寸的数据的支持;和能量功能，优化学习支持的概率质量。能量函数和基座都共同构成了最终模型，与GANS不同，它仅保留基本分布（“发电机”）。通过在学习能量和基础之间交替进行培训GEBMS。我们表明，两种培训阶段都明确定义：通过最大化广义可能性来学习能量，并且由此产生的能源的损失提供了学习基础的信息梯度。可以通过MCMC获得来自训练模型的潜在空间的后部的样品，从而在该空间中找到产生更好的质量样本的区域。经验上，图像生成任务上的GEBM样本比来自学习发电机的图像更好，表明所有其他相同，GEBM将优于同样复杂性的GAN。当使用归一化流作为基础测量时，GEBMS成功地启动密度建模任务，返回相当的性能以直接相同网络的最大可能性。

与CNN的分类，分割或对象检测相比，生成网络的目标和方法根本不同。最初，它们不是作为图像分析工具，而是生成自然看起来的图像。已经提出了对抗性训练范式来稳定生成方法，并已被证明是非常成功的 - 尽管绝不是第一次尝试。本章对生成对抗网络（GAN）的动机进行了基本介绍，并通​​过抽象基本任务和工作机制并得出了早期实用方法的困难来追溯其成功的道路。将显示进行更稳定的训练方法，也将显示出不良收敛及其原因的典型迹象。尽管本章侧重于用于图像生成和图像分析的gan，但对抗性训练范式本身并非特定于图像，并且在图像分析中也概括了任务。在将GAN与最近进入场景的进一步生成建模方法进行对比之前，将闻名图像语义分割和异常检测的架构示例。这将允许对限制的上下文化观点，但也可以对gans有好处。

We investigate the training and performance of generative adversarial networks using the Maximum Mean Discrepancy (MMD) as critic, termed MMD GANs. As our main theoretical contribution, we clarify the situation with bias in GAN loss functions raised by recent work: we show that gradient estimators used in the optimization process for both MMD GANs and Wasserstein GANs are unbiased, but learning a discriminator based on samples leads to biased gradients for the generator parameters. We also discuss the issue of kernel choice for the MMD critic, and characterize the kernel corresponding to the energy distance used for the Cramér GAN critic. Being an integral probability metric, the MMD benefits from training strategies recently developed for Wasserstein GANs. In experiments, the MMD GAN is able to employ a smaller critic network than the Wasserstein GAN, resulting in a simpler and faster-training algorithm with matching performance. We also propose an improved measure of GAN convergence, the Kernel Inception Distance, and show how to use it to dynamically adapt learning rates during GAN training.

在没有明确或易于处理的可能性的情况下，贝叶斯人经常诉诸于贝叶斯计算（ABC）进行推理。我们的工作基于生成的对抗网络（GAN）和对抗性变分贝叶斯（GAN），为ABC桥接了ABC。 ABC和GAN都比较了观察到的数据和假数据的各个方面，分别从后代和似然模拟。我们开发了一个贝叶斯gan（B-GAN）采样器，该采样器通过解决对抗性优化问题直接靶向后部。 B-GAN是由有条件gan在ABC参考上学习的确定性映射驱动的。一旦训练了映射，就可以通过以可忽略的额外费用过滤噪声来获得IID后样品。我们建议使用（1）数据驱动的提案和（2）变化贝叶斯提出两项后处理的本地改进。我们通过常见的bayesian结果支持我们的发现，表明对于某些神经网络发生器和歧视器，真实和近似后骨之间的典型总变化距离收敛到零。我们对模拟数据的发现相对于一些最新的无可能后验模拟器显示出竞争激烈的性能。

Generative adversarial networks (GANs) provide a way to learn deep representations without extensively annotated training data. They achieve this through deriving backpropagation signals through a competitive process involving a pair of networks. The representations that can be learned by GANs may be used in a variety of applications, including image synthesis, semantic image editing, style transfer, image super-resolution and classification. The aim of this review paper is to provide an overview of GANs for the signal processing community, drawing on familiar analogies and concepts where possible. In addition to identifying different methods for training and constructing GANs, we also point to remaining challenges in their theory and application.

为了稳定地训练生成对抗网络（GAN），将实例噪声注入歧视器的输入中被认为是理论上的声音解决方案，但是，在实践中尚未实现其承诺。本文介绍了采用高斯混合物分布的扩散 - 在正向扩散链的所有扩散步骤中定义，以注入实例噪声。从观察到或生成的数据扩散的混合物中的随机样品被作为歧视器的输入。通过将其梯度通过前向扩散链进行反向传播来更新，该链的长度可自适应地调节以控制每个训练步骤允许的最大噪声与数据比率。理论分析验证了所提出的扩散gan的声音，该扩散器提供了模型和域 - 不可分割的可区分增强。在各种数据集上进行的一系列实验表明，扩散 - GAN可以提供稳定且具有数据效率的GAN训练，从而使对强GAN基准的性能保持一致，以综合构成照片现实的图像。

基于似然或显式的深层生成模型使用神经网络来构建灵活的高维密度。该公式直接与歧管假设相矛盾，该假设指出，观察到的数据位于嵌入高维环境空间中的低维歧管上。在本文中，我们研究了在这种维度不匹配的情况下，最大可能的训练的病理。我们正式证明，在学习歧管本身而不是分布的情况下，可以实现堕落的优点，而我们称之为多种歧视的现象过于拟合。我们提出了一类两步程序，该过程包括降低降低步骤，然后进行最大样子密度估计，并证明它们在非参数方面恢复了数据生成分布，从而避免了多种歧视。我们还表明，这些过程能够对隐式模型（例如生成对抗网络）学到的流形进行密度估计，从而解决了这些模型的主要缺点。最近提出的几种方法是我们两步程序的实例。因此，我们统一，扩展和理论上证明了一大批模型。

许多生成模型通过使用确定性神经网络转换标准高斯随机变量来合成数据。这些模型包括变异自动编码器和生成对抗网络。在这项工作中，我们称它们为“推动”模型并研究其表现力。我们表明，这些生成网络的Lipschitz常数必须很大才能拟合多模式分布。更确切地说，我们表明，生成的和数据分布之间的总变化距离和kullback-leibler差异是根据模式分离和Lipschitz常数从下面界定的。由于限制神经网络的Lipschitz常数是稳定生成模型的一种常见方法，因此推动向前模型近似多模式分布的能力与训练的稳定性之间存在可证明的权衡。我们在一维和图像数据集上验证了我们的发现，并从经验上表明，在每个步骤中具有随机输入的堆叠网络组成的生成模型，例如扩散模型不会受到此类限制。

For distributions $\mathbb{P}$ and $\mathbb{Q}$ with different supports or undefined densities, the divergence $\textrm{D}(\mathbb{P}||\mathbb{Q})$ may not exist. We define a Spread Divergence $\tilde{\textrm{D}}(\mathbb{P}||\mathbb{Q})$ on modified $\mathbb{P}$ and $\mathbb{Q}$ and describe sufficient conditions for the existence of such a divergence. We demonstrate how to maximize the discriminatory power of a given divergence by parameterizing and learning the spread. We also give examples of using a Spread Divergence to train implicit generative models, including linear models (Independent Components Analysis) and non-linear models (Deep Generative Networks).

Quantum machine learning (QML) has received increasing attention due to its potential to outperform classical machine learning methods in various problems. A subclass of QML methods is quantum generative adversarial networks (QGANs) which have been studied as a quantum counterpart of classical GANs widely used in image manipulation and generation tasks. The existing work on QGANs is still limited to small-scale proof-of-concept examples based on images with significant down-scaling. Here we integrate classical and quantum techniques to propose a new hybrid quantum-classical GAN framework. We demonstrate its superior learning capabilities by generating $28 \times 28$ pixels grey-scale images without dimensionality reduction or classical pre/post-processing on multiple classes of the standard MNIST and Fashion MNIST datasets, which achieves comparable results to classical frameworks with 3 orders of magnitude less trainable generator parameters. To gain further insight into the working of our hybrid approach, we systematically explore the impact of its parameter space by varying the number of qubits, the size of image patches, the number of layers in the generator, the shape of the patches and the choice of prior distribution. Our results show that increasing the quantum generator size generally improves the learning capability of the network. The developed framework provides a foundation for future design of QGANs with optimal parameter set tailored for complex image generation tasks.

标准化流动，扩散归一化流量和变形自动置换器是强大的生成模型。在本文中，我们提供了一个统一的框架来通过马尔可夫链处理这些方法。实际上，我们考虑随机标准化流量作为一对马尔可夫链，满足一些属性，并表明许多用于数据生成的最先进模型适合该框架。马尔可夫链的观点使我们能够将确定性层作为可逆的神经网络和随机层作为大都会加速层，Langevin层和变形自身偏移，以数学上的声音方式。除了具有Langevin层的密度的层，扩散层或变形自身形式，也可以处理与确定性层或大都会加热器层没有密度的层。因此，我们的框架建立了一个有用的数学工具来结合各种方法。

生成对抗网络（GAN）在数据生成方面取得了巨大成功。但是，其统计特性尚未完全理解。在本文中，我们考虑了GAN的一般$ f $ divergence公式的统计行为，其中包括Kullback- Leibler Divergence与最大似然原理密切相关。我们表明，对于正确指定的参数生成模型，在适当的规律性条件下，所有具有相同歧视类别类别的$ f $ divergence gans均在渐近上等效。 Moreover, with an appropriately chosen local discriminator, they become equivalent to the maximum likelihood estimate asymptotically.对于被误解的生成模型，具有不同$ f $ -Divergences {收敛到不同估计器}的gan，因此无法直接比较。但是，结果表明，对于某些常用的$ f $ -Diverences，原始的$ f $ gan并不是最佳的，因为当更换原始$ f $ gan配方中的判别器培训时，可以实现较小的渐近方差通过逻辑回归。结果估计方法称为对抗梯度估计（年龄）。提供了实证研究来支持该理论，并证明了年龄的优势，而不是模型错误的原始$ f $ gans。

在概率密度范围内相对于Wassersein度量的空间的梯度流程通常具有很好的特性，并且已在几种机器学习应用中使用。计算Wasserstein梯度流量的标准方法是有限差异，使网格上的基础空间离散，并且不可扩展。在这项工作中，我们提出了一种可扩展的近端梯度型算法，用于Wassersein梯度流。我们的方法的关键是目标函数的变分形式，这使得可以通过引流 - 双重优化实现JKO近端地图。可以通过替代地更新内部和外环中的参数来有效地解决该原始问题。我们的框架涵盖了包括热方程和多孔介质方程的所有经典Wasserstein梯度流。我们展示了若干数值示例的算法的性能和可扩展性。

The modeling of probability distributions, specifically generative modeling and density estimation, has become an immensely popular subject in recent years by virtue of its outstanding performance on sophisticated data such as images and texts. Nevertheless, a theoretical understanding of its success is still incomplete. One mystery is the paradox between memorization and generalization: In theory, the model is trained to be exactly the same as the empirical distribution of the finite samples, whereas in practice, the trained model can generate new samples or estimate the likelihood of unseen samples. Likewise, the overwhelming diversity of distribution learning models calls for a unified perspective on this subject. This paper provides a mathematical framework such that all the well-known models can be derived based on simple principles. To demonstrate its efficacy, we present a survey of our results on the approximation error, training error and generalization error of these models, which can all be established based on this framework. In particular, the aforementioned paradox is resolved by proving that these models enjoy implicit regularization during training, so that the generalization error at early-stopping avoids the curse of dimensionality. Furthermore, we provide some new results on landscape analysis and the mode collapse phenomenon.

One of the challenges in the study of generative adversarial networks is the instability of its training. In this paper, we propose a novel weight normalization technique called spectral normalization to stabilize the training of the discriminator. Our new normalization technique is computationally light and easy to incorporate into existing implementations. We tested the efficacy of spectral normalization on CIFAR10, STL-10, and ILSVRC2012 dataset, and we experimentally confirmed that spectrally normalized GANs (SN-GANs) is capable of generating images of better or equal quality relative to the previous training stabilization techniques. The code with Chainer (Tokui et al., 2015), generated images and pretrained models are available at https://github.com/pfnet-research/sngan_ projection.

Learning high-dimensional distributions is often done with explicit likelihood modeling or implicit modeling via minimizing integral probability metrics (IPMs). In this paper, we expand this learning paradigm to stochastic orders, namely, the convex or Choquet order between probability measures. Towards this end, exploiting the relation between convex orders and optimal transport, we introduce the Choquet-Toland distance between probability measures, that can be used as a drop-in replacement for IPMs. We also introduce the Variational Dominance Criterion (VDC) to learn probability measures with dominance constraints, that encode the desired stochastic order between the learned measure and a known baseline. We analyze both quantities and show that they suffer from the curse of dimensionality and propose surrogates via input convex maxout networks (ICMNs), that enjoy parametric rates. We provide a min-max framework for learning with stochastic orders and validate it experimentally on synthetic and high-dimensional image generation, with promising results. Finally, our ICMNs class of convex functions and its derived Rademacher Complexity are of independent interest beyond their application in convex orders.

Recent work has shown local convergence of GAN training for absolutely continuous data and generator distributions. In this paper, we show that the requirement of absolute continuity is necessary: we describe a simple yet prototypical counterexample showing that in the more realistic case of distributions that are not absolutely continuous, unregularized GAN training is not always convergent. Furthermore, we discuss regularization strategies that were recently proposed to stabilize GAN training. Our analysis shows that GAN training with instance noise or zerocentered gradient penalties converges. On the other hand, we show that Wasserstein-GANs and WGAN-GP with a finite number of discriminator updates per generator update do not always converge to the equilibrium point. We discuss these results, leading us to a new explanation for the stability problems of GAN training. Based on our analysis, we extend our convergence results to more general GANs and prove local convergence for simplified gradient penalties even if the generator and data distributions lie on lower dimensional manifolds. We find these penalties to work well in practice and use them to learn highresolution generative image models for a variety of datasets with little hyperparameter tuning.