基于投影中心距离的支持向量预选取

Pre-extract support vectors based on projection center distance

支持向量机是一种能在训练样本数很少的情况下达到很好分类推广能力的学习算法.支持向量机在选择支撑矢量时却进行了大量不必要的运算,成为其应用的瓶颈问题.因此在基于支持向量的分类器学习算法中,预先选择支撑向量是非常重要的.投影中心距离算法是一种能够预选取支撑矢量的方法,该方法可以有效地预选取出包含支持向量的边界集,在不影响支持向量机的分类能力情况下,大大地减少了训练样本,提高了支持向量机的训练速度.本文采用投影中心距离算法进行支撑矢量的预选取,通过对人工线性、非线性数据及MINST字符库的实验证明了该方法的有效性和可行性.

Support vector machine is a satisfied classification, presents excellent performance in solving the problem with small sample. When Support vector machine choose support vector, it needs a lot of unnecessary computation. A new algorithm based on project center distance which can pre-extractging support vectors. The method greatly reduces the training samples and improves the speed of support vector machine,while the ability of support vector machine to classification is unaffected. Our experiments show that the algorithm can speed up the training process.