A Novel Adaptive Neural Network Regression Estimate Algorithm

Through combining the advantages of field theory with adaptive resonance theory and contraposing the characteristics of regression estimate problem, a novel neural network regression estimate algorithm FTART3 is proposed in this paper. This algorithm achieves fast learning speed and strong generalization ability. It not only has incremental learning ability but also overcomes the defect of manually configuring hidden neurons that is necessary for BP kind algorithms. Experiments including line, sine, 2D Mexican hat and 3D Mexican hat show that FTART3 is superior to BP kind algorithms, which are often used in regression estimate, on both function approximation effect and training time cost.