Improved Training of Wasserstein GANs

  • 类型:arxiv
  • 标识:1704.00028
  • 链接:https://arxiv.org/abs/1704.00028
  • 主题:llm-infra
  • 主分类:engineering
  • 形态:method
  • 被引:11129
  • 被引来源:Semantic Scholar
  • S2被引:11129
  • OpenAlex被引:1504
  • 影响力被引:1580
  • TLDR:This work proposes an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input, which performs better than standard WGAN and enables stable training of a wide variety of GAN architectures with almost no hyperparameter tuning.
  • OpenAlex ID:W2605135824
  • OpenAlex DOI:10.48550/arxiv.1704.00028
  • DOI:10.48550/arxiv.1704.00028
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://arxiv.org/pdf/1704.00028
  • OpenAlex更新:2026-08-24
  • 待LLM分类:否
  • 标题中文:Improved Training of Wasserstein GANs
  • TLDR中文:本文提出一种权重裁剪的替代方案:对 critic 相对于其输入的梯度范数施加惩罚。其性能优于标准 WGAN,能以几乎无需调参的方式稳定训练多种 GAN 架构。
  • 来源文件
  • [OpenAlex discover]
  • [S2 enrich]