arXiv：cs.LG· Katarina Petrovi\'c, Zander W. Blasingame, Danyal Rehman, \.Ismail \.Ilkan Ceylan, Michael Bronstein, Stephen Y. Zhang, Lazar Atanackovic, Alexander Tong·· 4 小时前AI 评分40
Global Transport：为无分类器引导流模型设计的全局传输耦合
Global Transport Couplings for Classifier-Free Guided Flows
AI 导读
研究提出 Global Transport（GT），一种不依赖类别标签的全局类别无关最优传输耦合，可让不同条件关联到源噪声的不同区域。单独使用时 GT 会拖累生成效果，但与无分类器引导（CFG）结合后，在离散类别与连续文本条件图像生成任务中，跨模型规模和采样预算均稳定提升生成质量。
正文
Abstract:Optimal-transport couplings have been shown to reduce training variance in unconditional flow models, but their role in conditional generation remains unclear. A natural approach constructs separate couplings for each condition, but this is impractical for large or continuous conditioning spaces found in modern image foundation models. We introduce Global Transport (GT), a global class-agnostic optimal-transport coupling, computed without class labels. GT can associate different conditions with different regions of the source noise, and consequently worsens performance without guidance. However, when combined with classifier-free guidance (CFG), GT consistently improves generation across domains, model scales, and sampling budgets. This reversal suggests that couplings for conditional flows should be evaluated both empirically and theoretically under the guided flow used at inference, rather than on unguided generation. We evaluate GT over both discrete class and continuous text conditioned image generation across model scales, and investigate how coupling choice alters guided trajectories. These results identify coupling design in the guided flow setting as a simple training time axis to improve performance without modifying existing architectures, samplers, or guidance mechanisms.
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.07555 [cs.LG] |
| (or arXiv:2610.07555v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.07555 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Katarina Petrović [view email]
[v1]
Tue, 6 Oct 2026 00:39:45 UTC (36,739 KB)
来源：arXiv：cs.LG · arxiv.org