介绍了支持向量机(SVM)的数学原理和最小二乘支持向量机(Least Squares Support Vector Machines,LSSVM)的数学原理与应用研究。在支持向量机中采用的是二次规划方法,而最小二乘支持向量机则用最小二乘线性系统作为损失函数从而取代它,...介绍了支持向量机(SVM)的数学原理和最小二乘支持向量机(Least Squares Support Vector Machines,LSSVM)的数学原理与应用研究。在支持向量机中采用的是二次规划方法,而最小二乘支持向量机则用最小二乘线性系统作为损失函数从而取代它,这样就利用等式约束的方法取代了不等式约束,最终演变为对线性方程组的求解,使求解的速度得到提高,求解的收敛精度得到提升。将最小二乘支持向量机与偏最小二乘法、标准支持向量机进行了对比。最终表明,LS-SVM计算结果更准确,更简单,内存的占有量也较少,计算时间短,耗时少,是一个很有应用价值的研究方向。展开更多
In this paper, a content based descriptor is pro- posed to retrieve 3D models, which employs histogram of local orientation （HLO） as a geometric property of the shape. The proposed 3D model descriptor scheme consist...In this paper, a content based descriptor is pro- posed to retrieve 3D models, which employs histogram of local orientation （HLO） as a geometric property of the shape. The proposed 3D model descriptor scheme consists of three steps. In the first step, Poisson equation is utilized to define a 3D model signature. Next, the local orientation is calculated for each voxel of the model using Hessian matrix. As the final step, a histogram-based 3D model descriptor is extracted by accumulating the values of the local orientation in bins. Due to efficiency of Poisson equation in describing the models with various structures, the proposed descriptor is capable of discriminating these models accurately. Since, the inner vox- els have a dominant contribution in the formation of the de- scriptor, sufficient robustness against noise can be achieved. This is because the noise mostly influences the boundary vox- els. Furthermore, we improve the retrieval performance us- ing support vector machine based one-shot score （SVM-OSS） similarity measure, which is more efficient than the conven- tional methods to compute the distance of feature vectors. The rotation normalization is performed employing the prin- cipal component analysis. To demonstrate the applicability of HLO, we implement experimental evaluations of precision- recall curve on ESB, PSB and WM-SHREC databases of 3D models. Experimental results validate the effectiveness of the proposed descriptor compared to some current methods.展开更多
Named entity disambiguation （NED） is the task of linking mentions of ambiguous entities to their referenced entities in a knowledge base such as Wikipedia. We propose an approach to effectively disentangle the discr...Named entity disambiguation （NED） is the task of linking mentions of ambiguous entities to their referenced entities in a knowledge base such as Wikipedia. We propose an approach to effectively disentangle the discriminative features in the manner of collaborative utilization of collective wisdom （via human-labeled crowd labels） and deep learning （via human-generated data） for the NED task. In particular, we devise a crowd model to elicit the underlying features （crowd features） from crowd labels that indicate a matching candidate for each mention, and then use the crowd features to fine-tune a dynamic convolutional neural network （DCNN）. The learned DCNN is employed to obtain deep crowd features to enhance traditional hand-crafted features for the NED task. The proposed method substantially benefits from the utilization of crowd knowledge （via crowd labels） into a generic deep learning for the NED task. Experimental analysis demonstrates that the proposed approach is superior to the traditional hand-crafted features when enough crowd labels are gathered.展开更多
支撑向量数据域描述(support vector data description,简称SVDD)作为一种已经得到广泛应用的核方法,目前研究主要集中在其性能和效率的提高上,然而该算法优化问题最优解性质的理论性质却没有得到足够的关注.为此,首先把SVDD定义的原始...支撑向量数据域描述(support vector data description,简称SVDD)作为一种已经得到广泛应用的核方法,目前研究主要集中在其性能和效率的提高上,然而该算法优化问题最优解性质的理论性质却没有得到足够的关注.为此,首先把SVDD定义的原始优化问题等价转化为一个凸约束二次优化问题,然后从理论上证明了其构建的超球圆心具有唯一性,然而超球半径在一定条件下却存在不唯一性,并且给出了半径存在不唯一性的充分必要条件.还从对偶优化问题的角度分析了超球的圆心和半径性质,并且给出了SVDD算法中在根据优化问题最优解构建超球半径不唯一情况下计算超球半径的方法.完善了该算法的理论和方法体系,从而为其更深入的研究和应用奠定了理论基础.展开更多
Network embedding aims at learning low-dimensional representation of vertexes in a network and effectively preserving network structures.These representations can be used as features for many complex tasks on networks...Network embedding aims at learning low-dimensional representation of vertexes in a network and effectively preserving network structures.These representations can be used as features for many complex tasks on networks such as community detection and multi-label classification.Some classic methods based on the skip-gram model have been proposed to learn the representation of vertexes.However,these methods do not consider the global structure(i.e.,community structure)while sampling vertex sequences in network.To solve this problem,we suggest a novel sampling method which takes community information into consideration.It first samples dense vertex sequences by taking advantage of modularity function and then learns vertex representation by using the skip-gram model.Experimental results on the tasks of community detection and multi-label classification show that our method outperforms three state-of-the-art methods on learning the vertex representations in networks.展开更多
文摘介绍了支持向量机(SVM)的数学原理和最小二乘支持向量机(Least Squares Support Vector Machines,LSSVM)的数学原理与应用研究。在支持向量机中采用的是二次规划方法,而最小二乘支持向量机则用最小二乘线性系统作为损失函数从而取代它,这样就利用等式约束的方法取代了不等式约束,最终演变为对线性方程组的求解,使求解的速度得到提高,求解的收敛精度得到提升。将最小二乘支持向量机与偏最小二乘法、标准支持向量机进行了对比。最终表明,LS-SVM计算结果更准确,更简单,内存的占有量也较少,计算时间短,耗时少,是一个很有应用价值的研究方向。
文摘In this paper, a content based descriptor is pro- posed to retrieve 3D models, which employs histogram of local orientation （HLO） as a geometric property of the shape. The proposed 3D model descriptor scheme consists of three steps. In the first step, Poisson equation is utilized to define a 3D model signature. Next, the local orientation is calculated for each voxel of the model using Hessian matrix. As the final step, a histogram-based 3D model descriptor is extracted by accumulating the values of the local orientation in bins. Due to efficiency of Poisson equation in describing the models with various structures, the proposed descriptor is capable of discriminating these models accurately. Since, the inner vox- els have a dominant contribution in the formation of the de- scriptor, sufficient robustness against noise can be achieved. This is because the noise mostly influences the boundary vox- els. Furthermore, we improve the retrieval performance us- ing support vector machine based one-shot score （SVM-OSS） similarity measure, which is more efficient than the conven- tional methods to compute the distance of feature vectors. The rotation normalization is performed employing the prin- cipal component analysis. To demonstrate the applicability of HLO, we implement experimental evaluations of precision- recall curve on ESB, PSB and WM-SHREC databases of 3D models. Experimental results validate the effectiveness of the proposed descriptor compared to some current methods.
基金supported by the National Basic Research Program of China(No.2015CB352300)the National Natural Science Foundation of China(Nos.61402401 and U1509206)+3 种基金the Zhejiang Provincial Natural Science Foundation of China(No.LQ14F010004)the China Knowledge Centre for Engineering Sciences and Technologythe Fundamental Research Funds for the Central Universitiesthe Qianjiang Talents Program of Zhejiang Province,China
文摘Named entity disambiguation （NED） is the task of linking mentions of ambiguous entities to their referenced entities in a knowledge base such as Wikipedia. We propose an approach to effectively disentangle the discriminative features in the manner of collaborative utilization of collective wisdom （via human-labeled crowd labels） and deep learning （via human-generated data） for the NED task. In particular, we devise a crowd model to elicit the underlying features （crowd features） from crowd labels that indicate a matching candidate for each mention, and then use the crowd features to fine-tune a dynamic convolutional neural network （DCNN）. The learned DCNN is employed to obtain deep crowd features to enhance traditional hand-crafted features for the NED task. The proposed method substantially benefits from the utilization of crowd knowledge （via crowd labels） into a generic deep learning for the NED task. Experimental analysis demonstrates that the proposed approach is superior to the traditional hand-crafted features when enough crowd labels are gathered.
文摘支撑向量数据域描述(support vector data description,简称SVDD)作为一种已经得到广泛应用的核方法,目前研究主要集中在其性能和效率的提高上,然而该算法优化问题最优解性质的理论性质却没有得到足够的关注.为此,首先把SVDD定义的原始优化问题等价转化为一个凸约束二次优化问题,然后从理论上证明了其构建的超球圆心具有唯一性,然而超球半径在一定条件下却存在不唯一性,并且给出了半径存在不唯一性的充分必要条件.还从对偶优化问题的角度分析了超球的圆心和半径性质,并且给出了SVDD算法中在根据优化问题最优解构建超球半径不唯一情况下计算超球半径的方法.完善了该算法的理论和方法体系,从而为其更深入的研究和应用奠定了理论基础.
基金Project supported by the National Natural Science Foundation of China(Grant No.61673085)the Program from the Sichuan Provincial Science and Technology,China(Grant No.2018RZ0081)the Fundamental Research Funds of China West Normal University(Grant No.17E063)。
文摘Network embedding aims at learning low-dimensional representation of vertexes in a network and effectively preserving network structures.These representations can be used as features for many complex tasks on networks such as community detection and multi-label classification.Some classic methods based on the skip-gram model have been proposed to learn the representation of vertexes.However,these methods do not consider the global structure(i.e.,community structure)while sampling vertex sequences in network.To solve this problem,we suggest a novel sampling method which takes community information into consideration.It first samples dense vertex sequences by taking advantage of modularity function and then learns vertex representation by using the skip-gram model.Experimental results on the tasks of community detection and multi-label classification show that our method outperforms three state-of-the-art methods on learning the vertex representations in networks.
