基于最小二乘法的标记分布学习

发布时间：2018-04-04 04:29

本文选题：标记分布　切入点：最小二乘　出处：《郑州大学学报(理学版)》2017年04期

【摘要】：多标记学习在一定程度上解决了标记多义性问题,它主要关注实例对应的相关标记或者无关标记,而标记分布能够反映相关标记对于实例的重要程度.从重构标记分布的思想出发,利用最小二乘法建立模型,提出基于最小二乘法的标记分布学习(lsm-LDL).首先用特征重构标记,通过变换矩阵使得每一个标记能够表示为特征的一个线性组合;然后用最小二乘法建立优化模型;最后引入L_2范数规则化项,防止过拟合,保证泛化能力.在4个实际的数据集上进行实验,并与3种已有的标记分布学习算法在5种评价指标上进行比较,实验结果表明提出的lsm-LDL算法是有效的.
[Abstract]:Multi-label learning solves the problem of label polysemy to a certain extent. It mainly focuses on the relative or irrelevant markers corresponding to the instance, and the distribution of markers can reflect the importance of the relevant markers to the instance.Based on the idea of reconstructing the label distribution, the model is established by using the least square method, and the label distribution learning based on the least square method is proposed.First, the label is reconstructed by the feature, and each label can be expressed as a linear combination by the transformation matrix. Then, the optimization model is established by using the least square method. Finally, the regularization term of L2 norm is introduced to prevent over-fitting.Ensure generalization ability.Experiments are carried out on four actual data sets and compared with three existing marker distributed learning algorithms on five evaluation indexes. The experimental results show that the proposed lsm-LDL algorithm is effective.
【作者单位】：闽南师范大学粒计算重点实验室;
【基金】：国家自然科学基金项目(61379049,61379089) 陕西省教育厅专项科研项目(16JK2015)
【分类号】：TP181

【相似文献】