基于域理论的自适应谐振神经网络研究

Research of Field Theory Based Adaptive Resonance Neural Network

提出了一种基于域理论的自适应谐振神经网络算法 FTART,有机结合了自适应谐振理论和域理论的优势 ,以一种独特的方式解决了示例间冲突和分类区域的动态扩展 ,不仅不需要手工设置隐层神经元 ,可以还获得了较快的训练速度和较高的预测精度 .同时还提出了一种可以从训练好的 FTART网络中抽取可理解性好、精度高的符号规则的方法 ,即基于统计的产生测试法 .实验结果表明 。

In this paper, a Field Theory based adaptive resonance neural network algorithm FTART, which combines the advantages of Adaptive Resonance Theory and Field Theory, is proposed. FTART employs a unique approach to solve the conflicts between instances and extend classification regions dynamically. So that it does not need user to manually configure hidden units, and achieves fast training speed and high predictive accuracy. Moreover, a method named Statistic based Producing and Testing, which has the ability of extracting comprehensive and accurate symbolic rules from trained FTART, is proposed. Experimental results show that the symbolic rules extracted via this method can commendably describe the function of FTART.