arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2602.01338cs.LGmath.STstat.MLstat.TH
收藏

扩散模型与对数凹分布的高精度采样

High-accuracy sampling for diffusion models and log-concave distributions

Department of XXX, University of YYY, Location, Country(XXX系,YYY大学,地点,国家);School of ZZZ, Institute of WWW, Location, Country(ZZZ学院,WWW研究所,地点,国家)

Fan Chen, Sinho Chewi, Constantinos Daskalakis, Alexander Rakhlin

AI总结:

本文提出扩散模型采样算法,实现δ误差在polylog(1/δ)步内完成,改进了现有结果。针对数据内在维度d*和非均匀L-Lipschitz条件,提出复杂度为polylog(1/δ)的采样方法。

摘要:

我们提出扩散模型采样算法,可在polylog(1/δ)步内实现δ误差,前提是能访问到L^2范数下δ精度的分数估计。这比以往所有结果有指数级改进。具体而言,在最小数据假设下,复杂度为~O(d* polylog(1/δ)),其中d*是数据的内在维度。进一步,在非均匀L-Lipschitz条件下,复杂度降至~O(L polylog(1/δ))。我们的方法还为仅使用梯度评估的一般对数凹分布提供了首个polylog(1/δ)复杂度的采样器。

英文摘要:

We present algorithms for diffusion model sampling which obtain $δ$-error in $\mathrm{polylog}(1/δ)$ steps, given access to $\widetilde O(δ)$-accurate score estimates in $L^2$. This is an exponential improvement over all previous results. Specifically, under minimal data assumptions, the complexity is $\widetilde O(d_\star \mathrm{polylog}(1/δ))$ where $d_\star$ is the intrinsic dimension of the data. Further, under a non-uniform $L$-Lipschitz condition, the complexity reduces to $\widetilde O(L \mathrm{polylog}(1/δ))$. Our approach also yields the first $\mathrm{polylog}(1/δ)$ complexity sampler for general log-concave distributions using only gradient evaluations.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑