Stochastic Primal Dual Splitting Algorithms for Convex and Nonconvex Composite Optimization in Imaging

Attend a virtual seminar in the Fifth Imaging & Inverse Problems (IMAGINE) OneWorld SIAM-IS series featuring speaker Xiaoqun Zhang from Shanghai Jiao Tong University. Explore stochastic primal dual splitting algorithms for convex and nonconvex composite optimization in imaging. Delve into two classes of algorithms: the first combines stochastic gradient with primal dual fixed point method (PDFP) for convex linearly composite problems, while the second focuses on Alternating Direction Method of Multipliers (ADMM) for nonconvex composite problems. Learn about the convergence and effectiveness of these algorithms through examples in graphic Lasso, graphics logistic regressions, and image reconstruction. Gain insights into the advantages of SVRG-PDFP for large-scale image reconstruction problems, especially with limited high-performance computing resources. Understand the global convergence and convergence rate of ADMM combined with variance reduction gradient estimators under the Kurdyka-Lojasiewicz (KL) function assumption.