通过重缩放得分函数，灵活控制扩散模型采样多样性。
Temporal Score Rescaling for Temperature Sampling in Diffusion and Flow Models
- 用时间相关的得分重缩放实现局部温度调节。
- 无需微调，可提升深度估计与图像生成性能。
- 适用于各类模型，适合需要多样性的场景。
我们提出一种机制，用于调控去噪扩散模型和流匹配模型的采样多样性，使用户能够从比训练分布更尖锐或更宽泛的分布中采样。基于这些模型利用（学习到的）噪声数据分布得分函数进行采样的观察，我们证明对得分函数进行重缩放可有效控制‘局部’采样温度。该方法无需任何微调或训练策略修改，可应用于任意现成模型，并兼容确定性和随机采样器。我们在二维模拟数据上验证了框架有效性，随后在五个不同任务上的扩散模型中展示其应用：图像生成、姿态估计、深度预测、机器人操作和蛋白质设计。结果显示，该方法能实现更尖锐（或更平坦）的采样分布，在深度预测中提升对高概率深度估计的采样能力，而在图像生成中稍加展平分布可获得更好性能。
原文摘要 · Abstract (English)
We present a mechanism to steer the sampling diversity of denoising diffusion and flow matching models, allowing users to sample from a sharper or broader distribution than the training distribution. We build on the observation that these models leverage (learned) score functions of noisy data distributions for sampling and show that rescaling these allows one to effectively control a 'local' sampling temperature. Notably, this approach does not require any finetuning or alterations to training strategy, and can be applied to any off-the-shelf model and is compatible with both deterministic and stochastic samplers. We first validate our framework on toy 2D data, and then demonstrate its application for diffusion models trained across five disparate tasks -- image generation, pose estimation, depth prediction, robot manipulation, and protein design. We find that across these tasks, our approach allows sampling from sharper (or flatter) distributions, yielding performance gains e.g., depth prediction models benefit from sampling more likely depth estimates, whereas image generation models perform better when sampling a slightly flatter distribution.
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。