一种基于多目标的自组织神经网络学习方法

自组织神经网络又称为无教师指导学习网络，可以自动地从环境中学习、获取知识、从而具有较强的自适应能力。目前，自组织神经网络在图象理解、模式识别、智能机器人控制等领域得到越来广泛的应用。但是，由于目前大部分组织神经网络都采用单准则无教师指导学习方法，从而导致了神经网络学习效率低等问题，这在一程度上影响了自组织神经网络更加广泛的有效应用。为此，本文提出了一种基于模糊熵准则和误差平方和准则的多目标（准则）自组织神经网络学习算法，该算法可以克服单准则无教师指导学习方法所存在的局限性，实验结果表明：该算法是有效的，并且较其它自组织神经网络学习方法，无论在学习效率上，还是在网络优化上，都具有很大的优越性。

A Kind of Multi-Criteria Learning Method for Self-organizing Neural Networks

Self organizing neural networks, also called unsupervised learning networks, may learn from the environment automatically, and therefore, possess stronger ability to learn adaptively. At present, self organizing neural networks have gotten wider and wider applications in the fields of image understanding,pattern recognition and intelligent robots. But, because of the single criterion learning algorithm used in self organizing neural networks, the learning efficiency of self organizing neural networks is low, which has affected the wider application of self organizing neural networks. This paper presents a kind of multi criteria learning method based on fuzzy entropy criterion and error squared sum criterion, which may overcome, to some extent, the limitations of single criterion. Experiment results have shown its advantages and effectiveness in both learning efficiency and neural network optimization.