增量支持向量机算法研究

Research on Increasing Support Vector Machine Algorithms

在进行增量学习时,随着新增样本的不断加入,致使训练集规模不断扩大,消耗大量计算资源,寻优速度缓慢。在深入研究了支持向量分布的特点的基础上提出了分治加权增量支持向量机算法。该算法有效利用了广义KKT条件和中心距离比值,舍弃对后续训练影响不大的样本,得到边界支持向量集,对训练样本进行有效的淘汰。将所剩样本合并,进行加权处理,解决某些样本严重偏离所属的类别,对正常分布的样本不公平的问题。实验结果表明,该方法在保证分类精度的同时,能有效地提高训练速度。

When carred on the increase studies, along with additional sample＇s unceasing joined, cause the training regulations scale unceasingly expanding, consumption of massive calculation resources, and the optimization speed is slow. Propose the partitioning weighting increase support vector machines algorithm in the deep research support vector distributed characteristic＇s foundation. This algorithm has effectively used the generalized KKT condition and center distance ratio, discards to the training sample that is not big influence, ob- tains the boundary support vector collection, carries on the effective elimination to the training sample. Then merge remain the sample, carries on weighting processing to solve the question that the certain sample serious deviae respective category, regarding normal distribu- tion sample unfair question. The experimental result indicated that this method can guarantee classification precision, can raise the train- ing speed effectively at the same time.