Computing and Mathematical Sciences Colloquium

Beginning with several modern applied problems arising from areas like signal processing and communication, sparsity representation, and two-stage stochastic programs, we present a unified class of nonsmooth, difference-of-convex (dc) programs that constitutes the main focus of our lecture. Topics to be discussed include: (i) clarification of several kinds of stationary solutions and their relations; (ii) development of a convex-programming based algorithm for computing a directional stationary problem of a convex constrained dc program, and (ii) extensions of the algorithm in several directions including a randomized version for selecting the subproblems to be solved, and if time permits, dc constrained problems.