- 演讲人: 于露(山东大学,教授)
- 时间:2026年12月8日15:30
- 地点:浙江大学紫金港校区行政楼1312会议室
- 主办单位:浙江大学数据科学研究中心
摘要:Score-based diffusion models have emerged as a powerful framework for generative modeling, with score estimation as a central statistical bottleneck. However, existing theoretical guarantees largely focus on light-tailed distributions or impose restrictive assumptions such as compact support, limiting their applicability to heavy-tailed data encountered in practice. In this work, we study standard score-based diffusion models for heavy-tailed distributions in a Sobolev class with smoothness $\beta>0$, covering both exponential and polynomial tail decay characterized by a tail parameter $\gamma$. Based on kernel density estimation, we derive sharp minimax rates for score estimation, revealing a qualitative distinction between the two regimes: under exponential tails, the estimation rate matches the light-tailed benchmark up to polylogarithmic factors, whereas under polynomial tails it explicitly depends on the tail parameter $\gamma$. We further establish the convergence guarantees for the associated continuous-time reverse diffusion dynamics. For exponential tails and $\beta\in(0,2]$, the generated distribution converges in total variation at the rate $n^{-\beta/(2\beta+d)}$ up to logarithmic factors, matching the light-tailed benchmark. For polynomial tails, we obtain a $\gamma$-dependent total variation bound, while leaving its optimality open. These results characterize the statistical limits of score estimation and the resulting sampling accuracy for heavy-tailed targets, extending diffusion theory beyond the light-tailed setting.
个人介绍:于露,山东大学数据科学研究院教授,入选国家级青年人才计划。曾在法国国立经济与统计研究中心(CREST/ENSAE Paris)从事博士后研究。获加拿大多伦多大学统计学博士学位、美国华盛顿大学统计学硕士学位,本科毕业于山东大学数学与应用数学专业。
主要研究方向为统计学与人工智能的数学理论,致力于研究数据科学与人工智能中的基础理论问题,当前重点关注生成式人工智能的理论基础。相关研究成果发表于 Journal of Econometrics、Bernoulli、NeurIPS、COLT、ICLR 等国际期刊和会议。