arXiv：cs.LG· Xuanchen Wang, Heng Wang, Weidong Cai·· 3 小时前AI 评分43
通过控制极坐标暴露延迟扩散模型记忆的 QGD 方法
Controlling Polar Exposure to Delay Memorization in Diffusion Models
AI 导读
研究者提出 Quality-Gated De-whitening（QGD），通过保留快速极坐标更新前缀并逐步恢复固定增益动量，延迟扩散模型的样本记忆。在 2,000 张 CIFAR-10 子集上，QGD 将极坐标基线的有效区间扩大 8.32 倍，常见检查点复制减少 75.9%，FID 为 75.56（SGD 为 79.37）。
正文
Abstract:Diffusion models can reach useful sample quality before copying training examples, but fast optimization can compress this generalization window by accelerating sample-specific fitting. We investigate this effect through update geometry and propose Quality-Gated De-whitening (QGD), a controller that retains a fast polar-update prefix and progressively restores fixed-gain momentum. Our random-feature analysis separates covariance-controlled, curvature-equalized and amplitude-controlled memorization clocks. Under aligned spectral assumptions, it establishes a finite-exposure condition under which a fixed-gain tail recovers a delay proportional to dataset size. QGD implements this principle with a confirmed quality gate, a bounded decay envelope and causal copy feedback. Immediate switching is the conservative limit; gradual control balances delayed copying against continued quality improvement. We pair QGD with Copy-Budgeted Selection (CBS), which applies simultaneous binomial calibration to a frozen checkpoint family, followed by a fresh evaluation of the released checkpoint. On 2,000-image CIFAR-10 subsets, QGD preserves the polar baseline's quality-arrival time while expanding its useful interval by 8.32x and reducing common-checkpoint copying by 75.9%. With identical calibration and independent quality evaluation, QGD achieves FID 75.56 versus 79.37 for SGD with the same selector. Exposure-matched controls, independent detector audits and transfer to flow matching and dance generation support adaptive exposure control as a practical way to improve the quality-copying tradeoff.
| Comments: | 31 pages, 4 figures |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.02780 [cs.LG] |
| (or arXiv:2610.02780v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.02780 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Xuanchen Wang [view email]
[v1]
Fri, 2 Oct 2026 04:11:59 UTC (330 KB)
来源：arXiv：cs.LG · arxiv.org