主講人：包承龍 清華大學(xué)長(zhǎng)聘副教授

時(shí)間：2025年5月25日10：00

地點(diǎn)：三號(hào)樓332室

舉辦單位：數(shù)理學(xué)院

主講人介紹：包承龍，清華大學(xué)丘成桐數(shù)學(xué)科學(xué)中心長(zhǎng)聘副教授、北京雁棲湖應(yīng)用數(shù)學(xué)研究院副研究員、清華大學(xué)膜生物學(xué)全國(guó)重點(diǎn)實(shí)驗(yàn)室研究員。2014 年博士畢業(yè)于新加坡國(guó)立大學(xué)數(shù)學(xué)系， 2015 年至 2018 年在新加坡國(guó)立大學(xué)數(shù)學(xué)系進(jìn)行博士后研究。研究興趣主要在圖像處理的建模與大規(guī)模優(yōu)化算法方面，擔(dān)任SIAM Journal on Imaging Sciences編委，已在各類(lèi)期刊和會(huì)議上發(fā)表學(xué)術(shù)論文50余篇。

內(nèi)容介紹：A significant gap between theory and practice in imaging sciences arises from inaccuracies in mathematical models, including imperfect imaging models and complex noise. Recent advancements have seen deep neural networks directly mapping observed data to clean images using paired training data. While these approaches deliver promising results across various tasks, collecting paired training data remains challenging and resource-intensive in practice. To address this limitation, we propose a unified generative model capable of leveraging both paired and unpaired data during training. Once trained, the model can generate high-quality synthetic data for direct use in downstream tasks. Experimental results on diverse real-world datasets demonstrate the effectiveness of the proposed methods. Finally, I will present recent progress in addressing the preferred orientation problem in cryo-EM, showcasing how these tools contribute to advancing the field.