Feng Yu, Yi Yang, Teng Zhang
The paper introduces a new ADMM-based algorithm for the graph-fused lasso that decomposes the objective into two parts to reduce per-iteration cost and achieve faster convergence.