Title: A novel ADMM algorithm for graph-fused lasso
Authors: Teng Zhang - University of Central Florida (United States) [presenting]
Abstract: A new algorithm is proposed for solving the graph-fused lasso (GFL), a method for parameter estimation that operates under the assumption that the signal tends to be locally constant over a predefined graph structure. The proposed method applies alternating direction method of multiplier (ADMM), which is based on the decomposition of the objective function into two components. While ADMM has been widely used in this problem, existing works decompose the objective function into the loss function component and the total variation penalty component. In comparison, the objective function is proposed to be decomposed into two parts, where one part is the loss function with some total variation penalty, and the other part is the remaining total variation penalty. Compared with existing works, this method has a smaller computational cost per iteration and fewer iterations to convergence in many settings. Experiments on artificial and real data sets confirm the competitive performance of our method.