基于疏密度约简的增量SVDD方法
首发时间:2020-01-23
摘要:针对增量支持向量数据描述方法存在计算复杂度高的问题,提出一种基于样本疏密度约简的增量学习方法。首先,从增量样本集中筛选出违反KKT条件的样本子集,并评价其整体聚集程度;然后,依据信息熵计算样本的疏密度值,结合比例控制因子再次筛选剩余样本集;最后,以约简的增量样本子集、初始模型的支持向量及边界附近样本作为数据集,增量学习得到新的识别模型。基于实测的风电机组叶片数据集,实验验证了方法的有效性。
For information in English, please click here
Incremental SVDD method based on density reduction
Abstract:To solve the problem of high computational complexity in the incremental support vector data description method, a novel incremental learning method based on sample density reduction is proposed in this paper. The aggregation degree of selected samples, which violate the KKT condition in incremental dataset, is first evaluated. Then the residual incremental samples are further reduced according to the information entropy and proportion control factor. Finally, the mixed dataset, consisting of reduced incremental samples, support vectors and those samples near decision boundary, is used to train new model. The validity of proposed method is verified by the dataset collected from wind turbine blades.
Keywords: support vector data description incremental learning density reduction
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于疏密度约简的增量SVDD方法
评论
全部评论0/1000