成为噪声是什么体验？一种面向扩散模型的基于熵的高斯噪声正则化
What Is It Like to Be a Noise? An Entropy-based Gaussian Noise Regularization for Diffusion Models
为扩散模型潜变量的推理时优化提供一个可单样本计算的「高斯性」正则项，可直接插入现有 latent optimization / 引导式编辑流程，缓解伪影与奖励黑客问题。
原文依据 we propose a Gaussianity regularizer that aligns a sample's local statistics with a typical Gaussian realization
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论对扩散潜变量进行推理时优化可实现强大的控制能力，但往往会破坏真实高斯噪声的统计结构，从而产生伪影与奖励黑客现象。为解决这一问题，我们提出一种高斯性正则项，使样本的局部统计量与一个典型的高斯实现相对齐，而非依赖逐点似然。我们通过计算样本分布与高斯先验之间的 KL 散度来形式化这一思想。为使该散度能够从单个样本出发可计算，我们将每个候选潜变量提升为由其统计量所诱导的经验分布，并将其建模为成对马尔可夫随机场。由此得到一个 Bethe-Kikuchi 型正则项，包含一维边缘项、二维空间项与多尺度项。结果表明，相较于已有方法，我们的方法提升了潜变量优化的稳定性与生成质量。
英文原文摘要 864 字符
Inference-time optimization of diffusion latents enables powerful control but often degrades the statistical structure of true Gaussian noise, causing artifacts and reward hacking. To address this, we propose a Gaussianity regularizer that aligns a sample's local statistics with a typical Gaussian realization, rather than relying on pointwise likelihood. We formalize this by computing the KL divergence between the sample distribution and the Gaussian prior. To make the divergence computation tractable from a single sample, we lift each candidate latent into an empirical distribution induced by its statistics and model it as a pairwise Markov Random Field. This yields a Bethe-Kikuchi-style regularizer with 1D marginal, 2D spatial, and multi-scale terms. Our results show improved latent optimization stability and generation quality over prior approaches.