A novel online incremental and decremental learning algorithm based on variable support vector machine
A novel online incremental and decremental learning algorithm based on variable support vector machine
复制标题
DOI：
10.1007/s10586-018-1772-4
复制
发表时间：
2019-05-01
影响因子：
4.4
通讯作者：
Zuo, Jingwen
中科院分区：
文献类型：
--
作者：
Chen, Yuantao;Xiong, Jie;Zuo, Jingwen
针对支持向量机在大规模训练样本中执行时间长、执行效率低的问题，提出了基于可变支持向量机的在线增减学习算法。在深入了解支持向量机的运行机制和相关算法的基础上，每个样本都增加了训练数据集的变化，需要更新分类器的学习算法。首先，给出在线增长量的学习算法，充分利用增量预先计算的信息，而不需要对新的增量训练数据集进行重新训练。其次，增量矩阵求逆过程大大减少了算法的运行时间，为了验证在线学习算法的有效性，给出了增量矩阵求逆过程。最后，在模式分类实验中选取了标准库中的九组数据集。实验结果表明，在给出的在线学习算法的情况下，保证了正确的分类率和有效的训练速度。随着增量过程的实施，训练会议的召开，需要大规模的数据存储空间，导致训练速度慢，基于VSVM的在线学习算法可以很好的解决这一问题。
In view of the long execution time and low execution efficiency of Support Vector Machine in large-scale training samples, the paper has proposed the online incremental and decremental learning algorithm based on variable support vector machine (VSVM). In deep understanding of the operation mechanism and correlation algorithms for VSVM, each sample has increased training datasets changes and it needs to update the classifier of learning algorithm. Firstly, they are given the online growth amount of learning algorithm taken full advantage of the incremental pre-calculated information, and doesn't require retraining for the new incremental training datasets. Secondly, the incremental matrix inverse calculation process had greatly reduced the running time of algorithm, and it is given in order to verify out the validity of the online learning algorithm. Finally, the nine groups of datasets in the standard library have been selected in the pattern classification experiment. The experimental results are shown that the online learning algorithm given in the case to ensure the correct classification rates and effective training's speed. With the implementation of the incremental process, training meetings, the need for large-scale data storage space, result in slow training, the online learning algorithm based on VSVM can solve the problem.