Abstract:
As the covering algorithm of forward neural networks design the network according to the training samples, its framework is easy to define and its performance is high. But the existence of noise samples may lead to some small covering domains, and these small domains increase the complexity and result in misrecognition. The support and confidence of a covering domain are defined and an algorithm of domains reduction is presented in this paper. Experimental results show that this algorithm can simplify the structure of the covering network and improve the stability of networks.
Key words:
neural networks,
covering domain,
reduction
摘要: 前向神经网络的网络覆盖算法根据样本数据构造性地建立神经网络,其结构易于确定,执行效率高。但由于噪声数据的存在,可能造成覆盖领域多的现象,增加了网络结构的复杂度,并产生一些不必要的误识。该文借鉴数据挖掘中关联规则的支持度与可信度的概念,对覆盖领域进行约简,理论分析和实验表明,该算法可以有效地简化覆盖网络的结构,提高网络的稳定性和推广能力。
关键词:
神经网络,
覆盖领域,
约简
CLC Number:
WU Tao; ; SHANG Li; WANG Wei; CHEN Li-wei. Algorithm of the Covering Domains Reduction Based on Association Rule[J]. Computer Engineering, 2008, 34(5): 57-59.
吴 涛; ;尚 丽;王 伟;陈黎伟. 基于关联规则的覆盖领域约简算法[J]. 计算机工程, 2008, 34(5): 57-59.