摘要
针对含单层专家网络的委员会机器在处理复杂问题时拟合能力不足的情况,本文提出了一种基于两层专家网络的委员会机器(CM-2LE)模型,并推导了其中的网络权值学习规则。对人造数据的整体检验和对实际的气象数据的逐次预报检验,通过调节隐含层节点数目,实验误差结果出现了较明显的减小过程,表明通过增加委员会机器中专家网络的层数,可以提高委员会机器的拟合能力。
A CM with two-layer expert net (CM-2LE) is presented to overcome the lack of approximating ability of CM with single layer of expert net for complicated problems. The derived learning rules of the CM-2LE is proposed. Experiments is performed on both synthetic and real-llfe climatic data. Results show that the increasing number of hidden layer of an expert net in the CM can improve the approximating ability of the CMs.
出处
《计算机科学》
CSCD
北大核心
2008年第1期160-163,180,共5页
Computer Science
基金
上海市科委项目(065115023)
华东师范大学211重点(521B0108)资助