A new recurrent neural network for solving convex quadratic programming problems with an application to the...

Area of Science:

• Computational mathematics
• Artificial intelligence
• Neural networks

Background:

• Quadratic programming (QP) and minimax problems are crucial in optimization.
• Existing neural network solutions often require strong conditions or complex computations like matrix inversion.

Purpose of the Study:

• To propose a new recurrent neural network for convex quadratic programming (QP) problems.
• To demonstrate its advantages over existing methods, including global convergence and structural simplicity.

Main Methods:

• Development of a novel recurrent neural network architecture.
• Variable substitution to connect the proposed network to existing minimax problem solvers.
• Design of a k-winners-take-all (k-WTA) network with O(n) complexity.

Main Results:

• The proposed network exhibits global convergence under weak conditions.
• It features low structural complexity and does not require matrix inverse calculations.
• The network is shown to be a special case of minimax neural networks and can solve ill-posed problems.

Conclusions:

• The new recurrent neural network is a competitive alternative for solving linear and quadratic programming problems.
• The network design methodology offers potential for future innovations in neural network-based optimization.
• Numerical simulations confirm the theoretical findings and practical applicability.