Neural Network Approximation of Nonlinear Noisy Functions Based on Coevolutionary Cooperative-Competitive Approach

An evolutionary algorithm for approximating nonlinear noisy functions based on coevolutionary models of cooperation and competition is proposed. This algorithm implements an environment that is conductive to cooperation and competition of populations in which each individual is a feedforward neural network that solves a specific problem. It is proposed to use populations of universal approximators for the studied function approximation and to introduce an additional population of denoising autoencoders for a possible noise reduction. The simulation results confirm the effectiveness of the proposed method of nonlinear noisy functions approximation.