针对多语义图像在用户图像检索反馈过程中带来的困扰,SVM在图像多分类过程中分类器同等对待等问题,提出基于K-means和SVM一对一多分类的图像反馈检索优化算法KWOVOSVM (K-means and weighted one-versus-one support vector machine)。...针对多语义图像在用户图像检索反馈过程中带来的困扰,SVM在图像多分类过程中分类器同等对待等问题,提出基于K-means和SVM一对一多分类的图像反馈检索优化算法KWOVOSVM (K-means and weighted one-versus-one support vector machine)。运用K-means算法对图像特征进行多次聚类,选取最具代表的信息图像样本供用户反馈;在用户反馈过程中,对其图像样本进行多分类训练时,通过欧式距离计算对每个分类器分配相对权重,使用户反馈次数减少,图像检索结果不断接近用户需求。实验结果表明,KWOVOSVM算法在查准率和满意度上有一定的提高。展开更多
Topological phases in non-Hermitian systems have become fascinating subjects recently.In this paper,we attempt to classify topological phases in 1D interacting non-Hermitian systems.We begin with the non-Hermitian gen...Topological phases in non-Hermitian systems have become fascinating subjects recently.In this paper,we attempt to classify topological phases in 1D interacting non-Hermitian systems.We begin with the non-Hermitian generalization of the Su-Schrieffer-Heeger(SSH)model and discuss its many-body topological Berry phase,which is well defined for all interacting quasi-Hermitian systems(non-Hermitian systems that have real energy spectrum).We then demonstrate that the classification of topological phases for quasi-Hermitian systems is exactly the same as their Hermitian counterparts.Finally,we construct the fixed point partition function for generic 1D interacting non-Hermitian local systems and find that the fixed point partition function still has a one-to-one correspondence to their Hermitian counterparts.Thus,we conclude that the classification of topological phases for generic 1D interacting non-Hermitian systems is still exactly the same as Hermitian systems.展开更多
文摘语种识别系统通常采用支持向量机(support vectormachine,SVM)一对多分类加Gauss后端分类器的方法进行分类。传统的SVM一对一分类在进行线性鉴别性分析(linear discriminant analysis,LDA)时特征值矩阵往往为奇异的,识别性能很低。该文提出一种改进的一对一分类方法,对SVM一对一分类得分重新建模,识别性能明显提高。在美国国家标准技术署(National Institute of Standardsand Technology,NIST)2011年语种识别评测(languagerecognition evaluation,LRE)30s数据集上的实验结果表明:在采用SVM的全变化量因子分析(total variability,iVector)和支持向量机-Gaussn超向量(support vectormachine-Gaussian super vector,SVM-GSV)语种识别系统上,该方法比SVM一对多分类方法性能更好,并且两种方法线性融合可明显提升识别性能,在iVector系统上各指标相对提升7.7%～15.9%,在SVM-GSV系统上各指标相对提升11.2%～33.9%。
文摘针对多语义图像在用户图像检索反馈过程中带来的困扰,SVM在图像多分类过程中分类器同等对待等问题,提出基于K-means和SVM一对一多分类的图像反馈检索优化算法KWOVOSVM (K-means and weighted one-versus-one support vector machine)。运用K-means算法对图像特征进行多次聚类,选取最具代表的信息图像样本供用户反馈;在用户反馈过程中,对其图像样本进行多分类训练时,通过欧式距离计算对每个分类器分配相对权重,使用户反馈次数减少,图像检索结果不断接近用户需求。实验结果表明,KWOVOSVM算法在查准率和满意度上有一定的提高。
基金supported by the National Key Research and Development Program of China (2016YFA0300300)the National Natural Science Foundation of China (NSFC+4 种基金11861161001)NSFC/RGC Joint Research Scheme (N-CUHK427/18)the Science, Technology and Innovation Commission of Shenzhen Municipality (ZDSYS20190902092905285)Guangdong Basic and Applied Basic Research Foundation under Grant No. 2020B1515120100Center for Computational Science and Engineering of Southern University of Science and Technology。
文摘Topological phases in non-Hermitian systems have become fascinating subjects recently.In this paper,we attempt to classify topological phases in 1D interacting non-Hermitian systems.We begin with the non-Hermitian generalization of the Su-Schrieffer-Heeger(SSH)model and discuss its many-body topological Berry phase,which is well defined for all interacting quasi-Hermitian systems(non-Hermitian systems that have real energy spectrum).We then demonstrate that the classification of topological phases for quasi-Hermitian systems is exactly the same as their Hermitian counterparts.Finally,we construct the fixed point partition function for generic 1D interacting non-Hermitian local systems and find that the fixed point partition function still has a one-to-one correspondence to their Hermitian counterparts.Thus,we conclude that the classification of topological phases for generic 1D interacting non-Hermitian systems is still exactly the same as Hermitian systems.
