题 目:Numerical Methods for Sparse Phase Retrieval
报 告 人:吕锡亮
单 位:武汉大学
时 间:2022年12月12日15:00
地 点:腾讯546-929-974
摘 要:In this talk we will present numerical algorithms for the sparse phase retrieval problem, which recovers an s-sparse signal $x\in\mathbb{R}^n$ from $m$ phaseless samples. The most existing sparse phase retrieval algorithms are first-order and hence converge at most linearly. We propose an efficient second-order algorithm for sparse phase retrieval by using the hard thresholding pursuit (HTP) algorithm in compressed sensing. The new proposed algorithm is theoretically guaranteed to give an exact sparse signal recovery in finite steps, in the Gaussian setting with $m=O(s log(n/s))$ and the a good initialization. Together with a spectral initialization, our algorithm is guaranteed to have an exact recovery from $O(s^2 logn)$ samples. Since the computational cost per iteration of our proposed algorithm is the same order as popular first-order algorithms, the algorithm is extremely efficient. Sample complexity can be further reduced numerically by introducing a stochastic alternating minimization strategy. Several experimental results show that the proposed algorithm can be several times faster than existing sparse phase retrieval algorithms.
报告人简介: 吕锡亮博士,武汉大学数学与统计学院教授。本科毕业于北京大学,并于新加坡国立大学获得硕士、博士学位;2007年1月至6月,赴美国马里兰大学做访问学者,2007年8月至2010年7月,在奥地利科学院RICAM研究所从事博士后研究;2010年起加入武汉大学数学与统计学院,任副教授、教授。2018年获长江学者青年学者(计算数学)。吕锡亮教授的研究方向为偏微分方程数值解、偏微分方程最优控制、反问题理论和计算、机器学习等。主持国家自然科学基金、科技部重点研发项目子课题、校企合作项目等,在计算数学的主流杂志如SINUM、SISC、Math. Comp.、Numer. Math.等发表论文五十多篇。