图像中存在的纹理、颜色和形状等异构视觉特征,在表示特定高层语义时所起作用的重要程度不同.为了在图像标注过程中更加有效地利用这些异构特征,提出了一种基于组稀疏(group sparsity)的多核学习方法(multiplekernel learning with grou...图像中存在的纹理、颜色和形状等异构视觉特征,在表示特定高层语义时所起作用的重要程度不同.为了在图像标注过程中更加有效地利用这些异构特征,提出了一种基于组稀疏(group sparsity)的多核学习方法(multiplekernel learning with group sparsity,简称MKLGS),为不同图像语义选择不同的组群特征.MKLGS先将包含多种异构特征的非线性图像数据映射到一个希尔伯特空间,然后利用希尔伯特空间中的核函数以及组LASSO(groupLASSO)对每个图像类别选择最具区别性特征的集合,最终训练得到分类模型对图像进行标注.通过与目前其他图像标注算法进行对比,实验结果表明,基于组稀疏的多核学习方法在图像标注中能取得很好的效果.展开更多
Considering the distinctiveness of different group features in the sparse representation, a novel joint multi- task and weighted group sparsity （JMT-WGS） method is pro- posed. By weighting popular group sparsity, no...Considering the distinctiveness of different group features in the sparse representation, a novel joint multi- task and weighted group sparsity （JMT-WGS） method is pro- posed. By weighting popular group sparsity, not only the rep- resentation coefficients from the same class over their asso- ciate dictionaries may share some similarity, but also the rep- resentation coefficients from different classes have enough di- versity. The proposed method is cast into a multi-task frame- work with two-stage iteration. In the first stage, representa- tion coefficient can be optimized by accelerated proximal gra- dient method when the weights are fixed. In the second stage, the weights are computed via the prior information about their entropy. The experimental results on three facial expres- sion databases show that the proposed algorithm outperforms other state-of-the-art algorithms and demonstrate the promis- ing performance of the proposed algorithm.展开更多
传统的CVaR条件风险价值组合投资模型能够很好的度量市场风险，但是容易在决策的过程中产生极端的投资权重，对CVaR模型增加一般范数约束后可以解决极端投资权重的问题，但却忽略了金融市场上常见的板块联动效应。基于上述原因，文章在...传统的CVaR条件风险价值组合投资模型能够很好的度量市场风险，但是容易在决策的过程中产生极端的投资权重，对CVaR模型增加一般范数约束后可以解决极端投资权重的问题，但却忽略了金融市场上常见的板块联动效应。基于上述原因，文章在传统的CVaR模型的基础上，施加Adaptive Group LASSO惩罚，构建了一种基于Adaptive Group LASSO的CVaR高维组合投资模型，通过Adaptive Group LASSO分位数回归求解算法，实现了在消除极端投资头寸的同时达到金融资产组水平上变量稀疏化的目的。最后，蒙特卡洛模拟与实证研究均发现，与传统的CVaR组合投资模型以及带有LAS—SO约束的CVaR组合投资模型相比，基于Adaptive Group LASSO的CVaR模型能够更好的考虑板块联动效应，并在行业组水平上选择相应的金融资产。展开更多
文摘图像中存在的纹理、颜色和形状等异构视觉特征,在表示特定高层语义时所起作用的重要程度不同.为了在图像标注过程中更加有效地利用这些异构特征,提出了一种基于组稀疏(group sparsity)的多核学习方法(multiplekernel learning with group sparsity,简称MKLGS),为不同图像语义选择不同的组群特征.MKLGS先将包含多种异构特征的非线性图像数据映射到一个希尔伯特空间,然后利用希尔伯特空间中的核函数以及组LASSO(groupLASSO)对每个图像类别选择最具区别性特征的集合,最终训练得到分类模型对图像进行标注.通过与目前其他图像标注算法进行对比,实验结果表明,基于组稀疏的多核学习方法在图像标注中能取得很好的效果.
基金This work was partially supported by the Project funded by China Postdoctoral Science Foundation （2014M5615556）, the National Natural Science Foundation of China （Grant Nos. 61273300, 61232007） and Jiangsu Natural Science Funds for Distinguished Young Scholar （BK20140022）. Also it is partially supported by Jiangsu Univer- sity Natural Science Funds （15KIB520024）, the State Key Laboratory for Novel Software Technology from Nanjing University （KFKT2014B18）, Jiangsu Key Laboratory of Image and Video Understanding for Social Safety （30920140122007）. Finally, the authors would like to thank the anonymous reviewers for their constructive advice.
文摘Considering the distinctiveness of different group features in the sparse representation, a novel joint multi- task and weighted group sparsity （JMT-WGS） method is pro- posed. By weighting popular group sparsity, not only the rep- resentation coefficients from the same class over their asso- ciate dictionaries may share some similarity, but also the rep- resentation coefficients from different classes have enough di- versity. The proposed method is cast into a multi-task frame- work with two-stage iteration. In the first stage, representa- tion coefficient can be optimized by accelerated proximal gra- dient method when the weights are fixed. In the second stage, the weights are computed via the prior information about their entropy. The experimental results on three facial expres- sion databases show that the proposed algorithm outperforms other state-of-the-art algorithms and demonstrate the promis- ing performance of the proposed algorithm.
文摘传统的CVaR条件风险价值组合投资模型能够很好的度量市场风险，但是容易在决策的过程中产生极端的投资权重，对CVaR模型增加一般范数约束后可以解决极端投资权重的问题，但却忽略了金融市场上常见的板块联动效应。基于上述原因，文章在传统的CVaR模型的基础上，施加Adaptive Group LASSO惩罚，构建了一种基于Adaptive Group LASSO的CVaR高维组合投资模型，通过Adaptive Group LASSO分位数回归求解算法，实现了在消除极端投资头寸的同时达到金融资产组水平上变量稀疏化的目的。最后，蒙特卡洛模拟与实证研究均发现，与传统的CVaR组合投资模型以及带有LAS—SO约束的CVaR组合投资模型相比，基于Adaptive Group LASSO的CVaR模型能够更好的考虑板块联动效应，并在行业组水平上选择相应的金融资产。
