A fixed- time neurodynamic approach for fused lasso problems

The fused lasso problem (FLP) arise from extensive applications such as biomedical engineering, signal processing and so on. However, the existing numerical algorithms are both time- and memory-inefficient due to its non-smoothness and non-separability. To overcome these shortcomings, Mohammadi recently proposed an efficient neurodynamic model for solving FLP (where named as FLSA). But the proposed neurodynamic approach is only guaranteed to converge globally to the optimal solution of the FLP. In this paper, we propose a neurodynamic approach with fixed-time convergence for FLP solving (named FxTNA for short). An explicit upper bound independent of the initial state on the time of convergence is given. Numerical simulations verify the FxTNA model's superiority in convergence performance and solution accuracy.