[09-15] 国防科技大学王红霞教授学术报告

• 报告题目：Title Fourier Phase Retrieval Under Deep Generative Prior

• 报告人：王红霞（国防科技大学数学系）

• 报告时间：2022年9月15日，星期四，上午10:00-11:00

• 报告地点：腾讯会议：929-256-258 密码：1123

• 报告摘要：Fourier phase retrieval (PR) problem is a nonlinear inverse problem arisen in various fields of science and engineering. Solving Fourier PR problem is challenging especially when the measurements are corrupted by noise. In this talk, we will focus on how to introduce deep generative prior to relieve the illness of Fourier PR.  Firstly, a new model called Data Driven Fourier Phase Retrieval (DFPR) together with the so-called DFPR-PnP ADMM algorithm are proposed by fully considering the nonconvex properties of Fourier PR and implicit prior learnt from data. Secondly, considering most images are essentially low-dimensional, or in the range of a deep generative neural network, the Deep-ADMM algorithm is proposed to solve Fourier PR problem constrained by a generative model. Under mild conditions, we show that the sequences generated by the Deep-ADMM can globally converge to the critical point. Numerous tests both on the simulated data and the optical measurements demonstrate the superiority of the newly proposed algorithms.

报告人简介：王红霞，国防科技大学理学院数学系教授，计算数学方向博士生导师，湖南省计算数学与应用软件学会副理事长。长期从事信息处理中的新型算法研究。主持国家自然科学基金项目4项、科技部重点研发项目课题、军队173重点项目课题等10项。指导研究生获部委级优秀学位论文奖2篇，获军队科技进步二等奖、国家教学成果二等奖和湖南省教学成果一等奖。

转载请注明来源

欢迎分享、点赞、在看🙂🙂🙂

如有需要推广的学术信息欢迎随时联系

识别下方二维码添加小编微信🤝🤝🤝