在聚类过程中,多视角数据的高维性和稀疏性使视角内描述样本的不同特征对聚类结果的影响不同,同一样本在不同的视角中对聚类的贡献也不同.层次化区分视角内不同特征的权重和相同样本在不同视角内的权重是提高多视角聚类性能的重要因素....在聚类过程中,多视角数据的高维性和稀疏性使视角内描述样本的不同特征对聚类结果的影响不同,同一样本在不同的视角中对聚类的贡献也不同.层次化区分视角内不同特征的权重和相同样本在不同视角内的权重是提高多视角聚类性能的重要因素.提出了具有特征级和样本级两级权重的多视角聚类(multi-view clustering based on two-level weights,MVC2W)算法.该算法引入了特征级和样本级注意力机制学习每个视角内不同特征的权重和每个样本在不同视角内的权重.两级注意力机制使算法在训练过程中能够更加关注重要的特征和重要的样本,更加合理地融合不同视角的信息,从而有效克服数据高维性和稀疏性对聚类结果的影响.此外,MVC2W将表征学习和聚类过程融为一体,协同训练、相互促进,进一步提升聚类性能.在5个稀疏程度不同的数据集上的实验结果表明:MVC2W算法的聚类性能比11个基线算法均有提升,尤其是在稀疏程度高的数据集上,MVC2W的提升更加显著.展开更多
We use the extrapolated Tikhonov regularization to deal with the ill-posed problem of 3D density inversion of gravity gradient data. The use of regularization parameters in the proposed method reduces the deviations b...We use the extrapolated Tikhonov regularization to deal with the ill-posed problem of 3D density inversion of gravity gradient data. The use of regularization parameters in the proposed method reduces the deviations between calculated and observed data. We also use the depth weighting function based on the eigenvector of gravity gradient tensor to eliminate undesired effects owing to the fast attenuation of the position function. Model data suggest that the extrapolated Tikhonov regularization in conjunction with the depth weighting function can effectively recover the 3D distribution of density anomalies. We conduct density inversion of gravity gradient data from the Australia Kauring test site and compare the inversion results with the published research results. The proposed inversion method can be used to obtain the 3D density distribution of underground anomalies.展开更多
Partition-based clustering with weighted feature is developed in the framework of shadowed sets. The objects in the core and boundary regions, generated by shadowed sets-based clustering, have different impact on the ...Partition-based clustering with weighted feature is developed in the framework of shadowed sets. The objects in the core and boundary regions, generated by shadowed sets-based clustering, have different impact on the prototype of each cluster. By integrating feature weights, a formula for weight calculation is introduced to the clustering algorithm. The selection of weight exponent is crucial for good result and the weights are updated iteratively with each partition of clusters. The convergence of the weighted algorithms is given, and the feasible cluster validity indices of data mining application are utilized. Experimental results on both synthetic and real-life numerical data with different feature weights demonstrate that the weighted algorithm is better than the other unweighted algorithms.展开更多
文摘在聚类过程中,多视角数据的高维性和稀疏性使视角内描述样本的不同特征对聚类结果的影响不同,同一样本在不同的视角中对聚类的贡献也不同.层次化区分视角内不同特征的权重和相同样本在不同视角内的权重是提高多视角聚类性能的重要因素.提出了具有特征级和样本级两级权重的多视角聚类(multi-view clustering based on two-level weights,MVC2W)算法.该算法引入了特征级和样本级注意力机制学习每个视角内不同特征的权重和每个样本在不同视角内的权重.两级注意力机制使算法在训练过程中能够更加关注重要的特征和重要的样本,更加合理地融合不同视角的信息,从而有效克服数据高维性和稀疏性对聚类结果的影响.此外,MVC2W将表征学习和聚类过程融为一体,协同训练、相互促进,进一步提升聚类性能.在5个稀疏程度不同的数据集上的实验结果表明:MVC2W算法的聚类性能比11个基线算法均有提升,尤其是在稀疏程度高的数据集上,MVC2W的提升更加显著.
基金supported by National major special equipment development(No.2011YQ120045)The National Natural Science Fund(No.41074050 and 41304023)
文摘We use the extrapolated Tikhonov regularization to deal with the ill-posed problem of 3D density inversion of gravity gradient data. The use of regularization parameters in the proposed method reduces the deviations between calculated and observed data. We also use the depth weighting function based on the eigenvector of gravity gradient tensor to eliminate undesired effects owing to the fast attenuation of the position function. Model data suggest that the extrapolated Tikhonov regularization in conjunction with the depth weighting function can effectively recover the 3D distribution of density anomalies. We conduct density inversion of gravity gradient data from the Australia Kauring test site and compare the inversion results with the published research results. The proposed inversion method can be used to obtain the 3D density distribution of underground anomalies.
基金Supported by the National Natural Science Foundation of China(61139002)~~
文摘Partition-based clustering with weighted feature is developed in the framework of shadowed sets. The objects in the core and boundary regions, generated by shadowed sets-based clustering, have different impact on the prototype of each cluster. By integrating feature weights, a formula for weight calculation is introduced to the clustering algorithm. The selection of weight exponent is crucial for good result and the weights are updated iteratively with each partition of clusters. The convergence of the weighted algorithms is given, and the feasible cluster validity indices of data mining application are utilized. Experimental results on both synthetic and real-life numerical data with different feature weights demonstrate that the weighted algorithm is better than the other unweighted algorithms.
