Title: "Cone-Free" infeasible-start primal-dual methods
Speaker: | Mehdi Karimi |
Affiliation: | University of Waterloo |
Room: | MC 6486 |
Abstract: We develop an infeasible-start primal-dual algorithm for convex optimization problems. Our approach is "cone-free" in the sense that we directly apply our techniques to the given "good" formulation without necessarily reformulating it in a conic form, however, our approach can also be applied to conic optimization problems. After defining our primal-dual central path, we introduce a long-step path following algorithm and prove that it solves the problem in polynomial time; returns an optimal solution if it exists, otherwise detects infeasibility or unboundedness. We introduce our Matlab-based code that solves a large class of problems including LP, SOCP, SDP, QCQP, Geometric programming, and Entropy programming among others.