论文部分内容阅读
针对标准SVM不能有效利用数据流形的局部信息以及对数据中的野值敏感的两点不足,提出一种基于自适应局部图嵌入加权罚SVM.算法在保持SVM优化框架不变的情况下,在目标函数中同时加入了对数据整体类间间隔最大化和数据局部流形分布的要求,优化了分类决策边界,简化了核化过程,同时在软间隔的样本惩罚系数中引入了数据的全局结构信息,增强了算法的鲁棒性.在人工、标准和图像数据集上的实验结果表明,所提出的方法是有效的.
Aiming at the problem that the standard SVM can not effectively use the local information of the data manifold and the two points which are sensitive to the outliers in the data, a weighted penalty SVM based on adaptive partial graph is proposed. While keeping the SVM optimization framework unchanged, In the objective function, the requirements of maximizing the interval between the data and the distribution of the data manifold are added simultaneously. The classification decision boundary is optimized and the coring process is simplified. At the same time, the global data is introduced into the sample penalty coefficient of the soft interval The structure information enhances the robustness of the algorithm.Experimental results on artificial, standard and image data sets show that the proposed method is effective.