摘要
结合Adaboost算法的加权投票机制,提高对传神经网络CPN(Counterpropagation Networks)的学习效率,提出新型快速分类算法(简称为ACPN).实验证明,新算法的学习最小误差比传统CPN算法下降了96%,训练时间同比下降44%,网络训练阶段误差下降趋势明显稳定.
A novel adaptive boosting theory-Counterpropagation Neural Network（ACPN） for solving forecasting problems is presented.The boosting concept is integrated into the CPN learning algorithm for learning effectively.Compared with traditional CPN,the minimal training error and learning time in ACPN network fell about 96% and 44%,respectively.Furthermore,the curve of trainning error in ACPN presents downtrend basically and has less fluctuation.
出处
《华南师范大学学报（自然科学版）》
CAS
北大核心
2011年第2期60-64,共5页
Journal of South China Normal University(Natural Science Edition)
关键词
对传神经网络
分类算法
推进学习
预测波动性
Counterpropagation Neural Network
classified algorithm
adaptive improved learning
forecasting volatility
