报告题目 Title:Analysis and Control of Diffusion Models: From Pretraining to Post-Training and Posterior Sampling

报告人 Speaker:庞彤瑶

报告人所在单位 Affiliation:清华大学

时间 Time:2026-11-25 14:30-15:30

地点 Venue:腾讯会议:159-307-655 会议密码:261114

报告摘要 Abstract:This talk studies diffusion models across the pipeline from pretraining to post-training and posterior sampling. We first analyze how the forward diffusion schedule affects generation accuracy and sampling complexity, providing guidance for the design of stronger base models. Given a pretrained model, we then consider two ways to steer its generative distribution toward specific objectives: reinforcement-learning post-training for reward alignment and direct posterior sampling for inverse problems. Together, these topics illustrate how diffusion models can be designed, analyzed, and controlled from base-model construction to downstream applications.

个人简介 Bio:庞彤瑶,现为清华大学丘成桐数学科学中心助理教授。2014年本科毕业于北京大学,2019年获新加坡国立大学数学博士学位。主要研究方向为深度生成模型的理论、算法及其在图像处理与反问题中的应用,重点关注扩散模型的采样与误差分析、强化学习后训练以及后验采样。

海报 Poster: 庞彤瑶 学术报告.jpg