Generative Adversarial Parallelization
Daniel Jiwoong Im AIFounded Inc. Toronto, ON、He Ma, Chris Dongjoo Kim and Graham W. Taylor University of Guelph Guelph, ON
Abstract
Generative Adversarial Networks (GAN) have become one of the most studied frameworks for unsupervised learning due to their intuitive formulation. They have also been shown to be capable of generating convincing examples in limited domains, such as low-resolution images. However, they still prove difficult to train in practice and tend to ignore modes of the data generating distribution. Quantitatively capturing effects such as mode coverage and more generally the quality of the generative model still remain elusive. We propose Generative Adversarial Parallelization (GAP), a framework in which many GANs or their variants are trained simultaneously, exchanging their discriminators. This eliminates the tight coupling between a generator and discriminator, leading to improved convergence and improved coverage of modes. We also propose an improved variant of the recently proposed Generative Adversarial Metric and show how it can score individual GANs or their collections under the GAP model.
中文速览
生成对抗网络（GAN）在训练时容易陷入模式崩溃（mode collapse），只盯着数据分布中的少数几个模式，同时生成器与判别器之间的紧耦合关系也让训练过程极不稳定。为此，作者提出了"生成对抗并行化"（Generative Adversarial Parallelization，GAP）框架：同时训练多个GAN，并在训练过程中周期性地随机交换各GAN的判别器，强迫每个生成器去欺骗来自不同GAN的判别器，从而打破生成器与判别器之间的过度适应关系。实验在合成数据集和真实图像数据集上均表明，GAP相比单独训练的GAN能覆盖更多数据模式、收敛更稳定；作者还改进了评估指标GAM，提出GAM-II来定量衡量GAP模型集合的整体生成质量。这项工作为解决GAN训练不稳定和模式缺失这两大核心痛点提供了一种简洁且易于与现有GAN变体结合的实用方案。
原文 arXiv:1612.04021;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1612.04021v1