摘要
基于域理论的自适应谐振算法FTART采用了独特的解决样本间冲突和动态扩大分类区域的方法,学习速度快、精度高,取得了很好的实用效果。但由于其学习到的知识是通过连接权隐式表示的,可理解性差,推理过程难以解释,使其进一步发展受到了很大限制。本文从功能性角度出发,提出了一种从FTART网络中抽取if-then规则的方法。实验结果表明,抽取出的if-then规则可理解性好、预测精度高,可以很好地描述FTART网络的性能。
The Field Theory based Adaptive Resonance Theory neural network algorithm FTART employs a unique conflict, resolution strategy and a method dynamically expanding classification area. Because of its fast speed and high accuracy of learning, FTART has been applied to many domains with success. However, the further progress of FTART is hindered because the knowledge learned by FTART is embedded in weights of the net, that is, implicit representation. Thus the comprehensibility is poor and the inference procedure is unexplainable. From the functionality point of view, we proposed a method that is able to extract if-then rules from trained FTART networks in this paper. Experimental result shows that those if-then rules are comprehensible, accurate, and describe the function of FTART commendably.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
1999年第4期381-385,共5页
Pattern Recognition and Artificial Intelligence
基金
国家自然科学基金
江苏省自然科学基金
