A fixed- time neurodynamic approach for fused lasso problems.

Area of Science:

• Computational mathematics
• Neurodynamic modeling
• Optimization algorithms

Background:

• The fused lasso problem (FLP) is crucial in fields like biomedical engineering and signal processing.
• Existing numerical algorithms for FLP are inefficient due to non-smoothness and non-separability.
• A prior neurodynamic approach (FLSA) offered global convergence but lacked guaranteed convergence time.

Purpose of the Study:

• To develop a novel neurodynamic approach for solving the fused lasso problem (FLP).
• To achieve fixed-time convergence for FLP, overcoming limitations of existing methods.
• To provide an explicit upper bound on convergence time, independent of initial conditions.

Main Methods:

• Development of a new fixed-time neurodynamic approach (FxTNA).
• Theoretical analysis to establish fixed-time convergence properties.
• Numerical simulations to compare FxTNA with existing methods.

Main Results:

• The proposed FxTNA model demonstrates fixed-time convergence for FLP.
• An explicit, initial-state-independent upper bound on convergence time is derived.
• Numerical simulations confirm FxTNA's superior convergence performance and solution accuracy.

Conclusions:

• The FxTNA model presents a significant advancement in solving the fused lasso problem.
• Fixed-time convergence offers predictable and efficient computation for FLP.
• FxTNA is a promising tool for applications requiring fast and accurate solutions to FLP.