摘要
提出一种基于高斯径向基函数网络的潜在剖面建模的方法,该方法利用高斯径向基函数网络与潜在剖面分析的相似性,对高斯径向基函数网络进行改变。对改变后的网络采用两步训练的方式,首先通过贝叶斯回归估计径向基参数,再使用传统后向传播算法进行训练,最后从整体的角度上得到从自变量到因变量的完整模型。该模型能在既保留潜在剖面分析的可解释性的情况下,同时得到较好的全局拟合优度。
Proposes a method of latent profile analysis based on Gauss radial basis function network. As the similarity between Gaoss radial basis function network and latent profile analysis, some changes has made on the traditional Gauss radial basis function network. A two-step training method is adopted for the altered network. First, the radial basis parameters al~ estimated by Bayes regression, and then the tradi- tional back propagation algorithm is used. Finally, gets the complete model from the independent variable to the dependent variable from the overall perspective. The model can not only retain the interpretability of the potential latent analysis, but also obtain better global good- ness of fit.
作者
黎鸣
LI Ming(College of Computer Science, Sichuan University, Chengdu 61006)
关键词
高斯径向基函数网络
潜在剖面分析
贝叶斯分类
Gauss Radial Basis Function Network
Latent Profile Analysis
Bayes Classification