摘要
根据Skutin提出的交叉验证理论,针对神经网络学习算法提出了神经网络稳定性的交叉验证模型,并选择4种应用广泛、具有代表性的神经网络作为研究对象,通过随机数据集和UCI数据集上的数据实验结果得出了BP、RBF、GRNN、ELM等4种神经网络的稳定性排序,并用统计检验方法对排序结果进行了检验。
According to cross-validation theory by Skutin,the cross-validation model of neural network stability is proposed. Four wildly used and representative neural networks are adopted as the subjects investigated and retrieved the rank of the stabilities of BP, RBF, GRNN, ELM, using the experiment results based on the random and UCI data sets.Finally, and the ranking is tested by the statistical method.
出处
《计算机工程与应用》
CSCD
北大核心
2010年第34期43-45,共3页
Computer Engineering and Applications
基金
国家自然科学基金(No.70911130228
No.70771067)~~
关键词
神经网络
稳定性
交叉验证
统计检验
neural network
stability
cross-validation
statistical test