A Neural Networks Based on Hermite Polynomial Functions
In this paper,a constructive one-hidden- layer network is introduced where each hidden unit employs a Hermite polynomial function for its activation function that is different from other units. This constructive Hermite polynomial network generally learns as effectively as,but generalizes much better than the conventional constructive one-hidden-layer networks.Specifically, both a structure level as well as a function level adaptation methodologies are utilized in constructing the network..The functional level adaptation scheme ensures that the growing or constructive network has different activation functions for each neuron such that the network may be able to capture the underlying input-output map more effectively.The activation functions considered consist of orthonormal Hermite polynomials. Several simulations for the regression problem are carried out to confirm the effectiveness and superiority of our proposed new algorithm. Applications to a simple two-category problem and a Cancer benchmark problem are also included to demonstrate the potential utility of our proposed algorithm to the classification problem. sigmoidal constructive ( its wide approximation never activation best generalization opportunities improve more units radial developed as paradigms, architecture. different varied a the functions
functional level adaptation Hermite polynomials incremental training algorithms
Rigui Zhou Duan Bin Aiqiu Nie
Department of Computer Science and Technology,Nanjing University of Aeronautics and Astronautics,Nan Fire-fighting Detachment,Dongying,Shandong,250014,China Jiangxi Provincial Communication Design Institute,Nanchang,P.R.China.330002
国际会议
北京
英文
2007-08-05（万方平台首次上网日期，不代表论文的发表时间）