用纠错编码改进的Mry支持向量机多类分类算法.docVIP

用纠错编码改进的M-ry支持向量机多类分类算法 摘要:针对m-ary支持向量机(svm)多类分类算法结构简单,但泛化能力较弱的特点,提出了与纠错编码理论相结合的改进的m-ary svm算法。首先,将原始类别信息编码作为信息码;然后结合纠错编码理论及期望的纠错能力,产生一定程度上性能最佳的编码,作为分类器训练的依据;最后,对于识别阶段输出编码中的错误分类利用检错纠错原理进行校正。实验结果表明,改进的算法通过引入尽可能少的冗余子分类器增强了标准m-ary svm多类分类算法的性能。 关键词:m-ary;支持向量机;纠错编码;多类分类;最小码间距离;输出校正码 enhanced m-ary support vector machine by error correction coding for multi-category classification 英文作者名bao jian, liu ran* 英文地址(school of computer science, hangzhou dianzi university, hangzhou zhejiang 310018, china) abstract: m-ary support vector machine (m-ary svm) for multi-category classification has the advantage of simple structure, but the disadvantage of weak generalization ability. this paper presented an enhanced m-ary svm algorithm in combination with error correction coding theory. the main idea of the approach was to generate a group of best codes based on information codes derived from the original category flags information, then utilize such codes as the basis for training the classifier, while in the final feed-forward phase the output codes composed of each sub-classifier could be corrected by error detection and correction principle if there exists any identifying error. the experimental results confirm the effectiveness of the improved algorithm brought about by introducing as few sub-classifiers as possible. key words: m-ary; support vector machine (svm); error correction coding; multi-category classification; minimum code distance; output correction code 0引言 作为机器学习实现方法之一的统计学习理论(statistical learning theory, slt),以及基于此理论的支持向量机[1-2](support vector machine, svm)凭借其显著的性能优势,近年来得到广大研究人员的关注,并已取得了大量的研究成果,而此前作为研究热点的人工神经网络(artificial neural networks, ann)虽然在工程实践中得到广泛的应用,但由于其建立在大数定理的渐近理论之上,要求学习样本足够多,且容易陷入局部极值或过学习的困境,并且在实际应用中隐含层的层数及每层神经元数目如何确定仍无规律可循,只能凭借使用者的经验进行实验试凑。正是由于神经网络自身存在的这些不足,导致了支持向量机算法研究的兴起。 标准支持向量机是针对两类分类问题提出的,根据有限样本信息在模型复杂性和学习能力之间寻求最佳折中,即通过最大化分类间隔得到最大的泛化能力,但是现实中的大多数模式识别问题都是多类分类问题,对此,通常的解决方法主要有4种:一类对一类(one-against-one, oao)、一类对其余(one-against-rest, oar)、有向无环图svm[3](decisio