具备跨源域学习能力的模糊聚类算法研究

发布时间：2018-05-07 05:09

本文选题：迁移学习 + 跨源域学习　；参考：《计算机应用研究》2017年02期

【摘要】：针对传统迁移学习聚类算法因单一源域到单一目标域且两者类别数必须一致的约束而达不到良好的聚类效果问题,提出了一种跨源域学习的聚类算法。该算法具有三大优点:a)仅扩大源域数目且取消了源域类别数的限定,算法可以自适应选择源域进行学习,所以算法的迁移学习能够得到较大的提升;b)由于所利用的源域知识不会暴露原数据,所以算法具有良好的源域数据隐私保护性;c)通过调节平衡参数可以使算法退化为传统的聚类算法,因此该算法的聚类性能是有所保障的。通过在模拟数据集和真实数据集上的实验,验证了该算法较之现有迁移学习聚类算法具有更好的迁移能力,且聚类性能及鲁棒性也有较大的提升。
[Abstract]:Aiming at the problem that the traditional migration learning clustering algorithm can not achieve good clustering effect due to the constraints of single source domain to single target domain and the number of categories must be consistent, a cross-source learning clustering algorithm is proposed. The algorithm only expands the number of source fields and cancels the number of source domain categories, so the algorithm can choose the source domain to learn. Therefore, the transfer learning of the algorithm can be greatly promoted.) since the source domain knowledge used does not expose the original data, Therefore, the algorithm has good privacy protection in source domain. By adjusting the balance parameters, the algorithm can be reduced to a traditional clustering algorithm, so the clustering performance of the algorithm is guaranteed. The experiments on simulated data sets and real data sets show that the proposed algorithm has better migration capability than the existing migration learning clustering algorithms, and the clustering performance and robustness are improved greatly.
【作者单位】：江南大学数字媒体学院;
【基金】：江苏省自然科学基金资助项目(BK2011003)
【分类号】：TP311.13

【相似文献】