摘要
针对Albus CMAC在学习精度与存贮容量之间的矛盾,借鉴神经网络集成思想,并引入可信度的概念,提出了基于信度分配的串行集成CMAC,以提高学习系统的泛化能力和网络收敛速度。通过对复杂非线性函数的逐级降维,分步逼近,有效地提高了网络的学习精度。仿真研究进一步验证了该方案的可行性和有效性。
In order to solve the conflict between accuracy and memory capability of Albus CMAC, the ideas of neural network ensemble and credit assignment were introduced, and series ensemble CMAC based on credit assignment was proposed. The new arithmetic could effectively improve the computation accuracy. Simulation results show the feasibility and validity of the improved arithmetic by approaching complex nonlinear function step by step.
出处
《华东理工大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2008年第5期751-754,共4页
Journal of East China University of Science and Technology
基金
国家自然科学基金项目(60774078)
