arXiv:2510.12636stat.MLcs.LG2025-10

用一维分位数函数自适应学习数据噪声,提升生成模型效果。

Adapting Noise to Data: Generative Flows from 1D Processes

  • 用一维分位数函数参数化噪声分布,可灵活适配各类数据。
  • 在气象与图像数据上验证,生成质量显著提升且开销极低。
  • 适合需要高效生成高质量样本的科研与工程场景。

流模型默认的高斯潜变量在学习重尾分布等复杂数据时存在挑战。本文提出一种通用框架,通过一维分位数函数学习数据自适应的参数化先验分布(潜噪声),并利用噪声与数据间的Wasserstein距离进行优化。该分位数先验自然适配重尾和紧支撑分布,同时缩短了传输路径。在重尾气象与图像数据集上的数值结果表明,该方法具有优异的灵活性与有效性,且计算开销几乎可忽略。

原文摘要 · Abstract (English)

The default Gaussian latent in flow-based generative models poses challenges when learning certain distributions such as heavy-tailed ones. We introduce a general framework for learning data-adaptive parametric prior distributions (latent noise) using one-dimensional quantile functions, optimized via the Wasserstein distance between noise and data. The quantile-based prior parameterization naturally adapts to both heavy-tailed and compactly supported distributions and shortens transport paths. Numerical results on heavy-tailed weather and image datasets confirm the method's flexibility and effectiveness achieved with negligible computational overhead.

生成模型流模型噪声学习

Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品;中文卡片由大模型生成,请以原文为准。