论文部分内容阅读
为了减少求支持向量过程中二次规划的复杂度,利用训练样本集的几何信息,选出两类中离另一类最近的边界向量集合,它是样本中最有可能成为支持向量的一部分,用它代替原样本集进行训练.对新增样本,若存在违反KKT条件的样本,只对这部分新样本进行学习.同时找出原样本中可能转化为支持向量的非支持向量样本.基于分析结果,提出了一种新的基于最近边界向量的增量式支持向量机学习算法.对标准数据集的实验结果表明,算法是可行的,有效的.
In order to reduce the complexity of the quadratic programming in the process of finding the support vectors, the nearest neighbor vector set of the two classes is selected by using the geometric information of the training sample set. It is the part of the sample that is most likely to be the support vector, Use it instead of the original sample set for training.For the new sample, if there is a sample that violates the KKT condition, only this part of the new sample will be studied.At the same time, find the unsupported vector samples that may be transformed into the support vector in the original sample. As a result, a new incremental support vector machine learning algorithm based on the nearest boundary vector is proposed.Experimental results on a standard dataset show that the algorithm is feasible and effective.