Tianyu Wu (吴天宇), University of California, Los Angeles


A Primal-Dual Forward-Backward Algorithm with Nonstandard Metrics


Many optimization problems can be formulated in a primal dual setting. In this talk we will focus on a forward backward splitting scheme applied to the KKT condition of the primal dual problem. With the help of nonstandard metrics, we can get algorithms with easy to solve subproblems. The proof of convergence is similar to the standard metric case. It will be illustrated that various choice of nonstandard metrics will lead to different algorithms. We will also show that many existing algorithms can be analyzed in this general scheme.

« Back