论文部分内容阅读
针对半监督学习中渐进直推支持向量机(PTSVM)算法每次标注的样本数太少、训练速度慢、回溯式学习多、学习性能不稳定的问题,提出一种快速的渐进直推支持向量机学习算法.该算法利用支持向量的信息,基于支持向量域描述(SVDD)选择新标注、无标签的样本点,以区域标注法代替 PTSVM 的成对标注法,不仅继承了其渐进赋值和动态调整的规则,而且在保持甚至提高算法精度的同时,大大提高算法速度.在人工模拟数据和真实数据上的实验结果表明该算法的有效性.
In order to solve the problem of too few samples, slow training speed, multiple tracing back learning and unstable learning performance, the algorithm of progressive direct support vector machine (PTSVM) in semi-supervised learning proposes a fast progressive direct support vector This algorithm uses the information of support vector to select the new label and unlabeled sample point based on SVDD instead of the pairwise labeling method of PTSVM by region annotation method, which not only inherits its progressive assignment and dynamic Adjust the rules, and greatly improve the speed of the algorithm while maintaining or even improving the accuracy of the algorithm.Experimental results on the artificial data and the real data show the effectiveness of the algorithm.