Abstract：We found that neuron model is inadequate owing to its defects such as those inherent in its structure and in its capability of information storage. So we propose an intelligent neurons assemblage model with generalized wavelet basis function network as its excited function. Not only the wavelet neural networks' convergence rate is much faster and its nonlinear approach capability is much better but also its intelligent characteristics, such as the variable-scale adaptive adjustment of structure and the generalized information storage, make it reflect much more faithfully the biological original. Static learning and pseudo dynamic learning are demonstrated to prove that the proposed mechanism is valid.
 Salatian A W, Keon young yi, Yuan F Zheng. Reinforcement Learning for a Biped Robot to Climb Sloping Surfaces.Journal of Robotic Systems, 1997,14(4):283-296.  Sejnowski T. KorchC, Churchland P. Computational Neuroscince. Science,1998, 241:1299-1306.  Qinghua Zhang. Using Wavelet Network in Nonparametric Estimation. IEEE Trans on Neural Networks. 1997,8(2).  张山.周凤岐.一种基于广义子波基函数选择的智能化神经元集聚模型.西北工业大学学报,1998,16(4):495 500.  Gullapalli V, Franklin J A, Benbrahim H. Acquiring Robot Skills Via Reinforcement Iearning. IEEE Control Syst Mag,1994. 13-24