报告主题：基于随机先验深度算子网络中的不确定量化

时间：2024年4月7日13:00-13:45

地点：上川路校区二教105/腾讯会议 会议号：302 214 942 密码：6666

报告人：王艺红

报告内容简介：

A new direction in scientific deep learning aims at approximating maps between spaces of functions opening the path to building fast surrogate models that can be trained and queried in a continuous fashion at any resolution. One prominent approach for supervised learning under this setting is the so-called deep operator network architecture (DeepONet).We give a brief review of deep learning techniques for uncertainty quantification and introduce the randomized prior method.Examplesare presentedto demonstrate the ability of UQDeepONets to provide robust output function predictions, return reliable uncertainty estimates, as well as their ability to identify model bias, out-of-distribution function samples, and handle noisy data.

主讲人简介：

王艺红，女，统计与数学学院教授。兼任中国商业经济学会经济数学与统计学研究分会常务理事，美国数学评论评论员。主持国家自然科学基金、上海市、浙江省自然科学基金项目等课题3项，参与国防基础科研核科学挑战专题“百亿亿次计算科学的计算方法与高效能实现”领域项目，在数学领域高质量科技期刊（SIAM Journal on Scientific Computing，Journal of Computational Physics，Journal of Scientific Computing等）发表科研论文20余篇，出版专著1部。