A collaborative neurodynamic approach with two-timescale projection neural networks designed via...

Area of Science:

• Artificial Intelligence
• Optimization Theory
• Neural Networks

Background:

• Nonconvex optimization problems are prevalent in various scientific and engineering fields.
• Distributed optimization presents significant computational challenges.
• Existing methods often struggle with global convergence for nonconvex problems.

Purpose of the Study:

• To propose novel two-timescale projection neural networks for nonconvex and distributed nonconvex optimization.
• To establish the global convergence properties of the proposed neural networks.
• To develop a collaborative neurodynamic approach for enhanced global optimization.

Main Methods:

• Development of two two-timescale projection neural networks.
• Application of the majorization-minimization principle.
• Utilizing a meta-heuristic rule for re-initialization in a collaborative neurodynamic framework.

Main Results:

• The proposed neural networks are proven to be globally convergent to Karush-Kuhn-Tucker points.
• The collaborative neurodynamic approach enhances performance in global and distributed optimization.
• Numerical examples validate the efficacy of the developed methods.

Conclusions:

• The introduced two-timescale projection neural networks provide a theoretically sound and practically effective approach for nonconvex optimization.
• The collaborative neurodynamic strategy offers a powerful tool for tackling complex distributed optimization tasks.
• This work advances the state-of-the-art in neural network-based optimization techniques.