发表机构
ETH Zürich; Disney Research | Studios（苏黎世联邦理工学院; 迪士尼研究院与工作室）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
提出对比噪声对齐（CNA），通过动态优化噪声表示并利用InfoNCE目标及正则化项，改善生成流中噪声与数据的耦合，减少流曲率，在少步生成中显著降低FID。
AI 中文摘要
扩散模型和流匹配模型通常通过独立采样的高斯噪声来破坏数据进行训练。虽然这种方法简单且可扩展，但其前向过程引入了任意的数据-噪声耦合，迫使网络学习不相关端点之间的高曲率传输。现有的最优传输方法通过将固定的噪声样本重新分配给数据来减轻这一负担，但源噪声分布本身仍然是被动的。为了解决这个问题，我们引入了对比噪声对齐（CNA），一种在训练时通过直接优化噪声表示来创建动态、对比耦合的方法。通过将噪声批次建模为相互作用的粒子系统，CNA采用跨模态的InfoNCE目标来将噪声粒子与其配对的数据目标对齐。为了防止空间坍缩，这种对齐通过角度熵项和径向范数惩罚进行正则化。我们在理论上证明了该平衡态渐近地保持高斯结构，在推理过程中保持可处理性。在实验上，CNA改善了噪声与数据之间的对齐，减少了流曲率，并以更少的采样步骤提供了更好的生成质量。对于少步、像素空间生成（2-4次神经函数评估），与标准整流流相比，CNA将FID降低了超过50%，并且比最优传输基线至少降低了24%。
英文摘要
Diffusion and flow-matching models are typically trained by corrupting data through independently sampled Gaussian noise. While simple and scalable, this forward process induces arbitrary data-noise couplings, forcing the network to learn high-curvature transports between unrelated endpoints. Existing optimal-transport methods reduce this burden by reassigning fixed noise samples to data, but the source noise distribution itself remains passive. To address this, we introduce Contrastive Noise Alignment (CNA), a training-time method that creates dynamic, contrastive couplings by optimizing the noise representations directly. By modeling the noise batch as an interacting particle system, CNA employs a cross-modal InfoNCE objective to align noise particles with their paired data targets. To prevent spatial collapse, this alignment is regularized using an angular entropy term and a radial norm penalty. We show theoretically that this equilibrium asymptotically preserves Gaussian structures, maintaining tractability during inference. Empirically, CNA improves the alignment between noise and data, reduces flow curvature, and provides better generation quality with fewer required sampling steps. For few-step, pixel-space generation (2-4 NFEs), CNA reduces FID by over 50\% compared to standard rectified flow, and by at least 24\% against Optimal Transport baselines.
Comments21 pages, 10 figures, 9 tables
