Qiaolin Ye

1-Norm least squares twin support vector machines

During the last few years, nonparallel plane classifiers, such as Multisurface Proximal Support Vector Machine via Generalized Eigenvalues (GEPSVM), and Least Squares TWSVM (LSTSVM), have attracted much attention. However, there are not any modifications of them that have been presented to automatically select the input features. This motivates the rush towards new classifiers. In this paper, we develop a new nonparallel plane classifier, which is designed for automatically selecting the relevant features. We first introduce a Tikhonov regularization (TR) term that is usually used for regularizing least squares into the LSTSVM learning framework, and then convert this formulation to a linear programming (LP) problem. By minimizing an exterior penalty (EP) problem of the dual of the LP formulation and using a fast generalized Newton algorithm, our method yields very sparse solutions, such that it generates a classifier that depends on only a smaller number of input features. In other words, this approach is capable of suppressing input features. This makes the classifier easier to store and faster to compute in the classification phase. Lastly, experiments on both toy and real problems disclose the effectiveness of our method.

Feature selection for least squares projection twin support vector machine

In this paper, we propose a new feature selection approach for the recently proposed Least Squares Projection Twin Support Vector Machine (LSPTSVM) for binary classification. 1-norm is used in our feature selection objective so that only non-zero elements in weight vectors will be chosen as selected features. Also, the Tikhonov regularization term is incorporated to the objective of our approach to reduce the singularity problems of Quadratic Programming Problems (QPPs), and then to minimize its 1-norm measure. This approach leads to a strong feature suppression capability, called as Feature Selection for Least Squares Projection Twin Support Vector Machine (FLSPTSVM). The solutions of FLSPTSVM can be obtained by solving two smaller QPPS arising from two primal QPPs as opposed to two dual ones in Twin Support Vector Machine (TWSVM). Thus, FLSPTSVM is capable of generating sparse solutions. This means that FLSPTSVM can reduce the number of input features for a linear case. Our linear FLSPTSVM can also be extended to a nonlinear case with the kernel trick. When a nonlinear classifier is used, the number of kernel functions required for the classifier is reduced. Our experiments on publicly available datasets demonstrate that our FLSPTSVM has comparable classification accuracy to that of LSPTSVM and obtains sparse solutions.

L1-Norm Distance Linear Discriminant Analysis Based on an Effective Iterative Algorithm

Recent works have proposed two L1-norm distance measure-based linear discriminant analysis (LDA) methods, L1-LD and LDA-L1, which aim to promote the robustness of the conventional LDA against outliers. In LDA-L1, a gradient ascending iterative algorithm is applied, which, however, suffers from the choice of stepwise. In L1-LDA, an alternating optimization strategy is proposed to overcome this problem. In this paper, however, we show that due to the use of this strategy, L1-LDA is accompanied with some serious problems that hinder the derivation of the optimal discrimination for data. Then, we propose an effective iterative framework to solve a general L1-norm minimization–maximization (minmax) problem. Based on the framework, we further develop a effective L1-norm distance-based LDA (called L1-ELDA) method. Theoretical insights into the convergence and effectiveness of our algorithm are provided and further verified by extensive experimental results on image databases.

Graph regularized multilayer concept factorization for data representation

Previous studies have demonstrated that matrix factorization techniques, such as Nonnegative Matrix Factorization (NMF) and Concept Factorization (CF), have yielded impressive results in image processing and data representation. However, conventional CF and its variants with single layer factorization fail to capture the intrinsic structure of data. In this paper, we propose a novel sequential factorization method, namely Graph regularized Multilayer Concept Factorization (GMCF) for clustering. GMCF is a multi-stage procedure, which decomposes the observation matrix iteratively in a number of layers. In addition, GMCF further incorporates graph Laplacian regularization in each layer to efficiently preserve the manifold structure of data. An efficient iterative updating scheme is developed for optimizing GMCF. The convergence of this algorithm is strictly proved; the computational complexity is detailedly analyzed. Extensive experiments demonstrate that GMCF owns the superiorities in terms of data representation and clustering performance.

Analysis of the Complete Mitochondrial Genome Sequence of the Diploid Cotton Gossypium raimondii by Comparative Genomics Approaches

Cotton is one of the most important economic crops and the primary source of natural fiber and is an important protein source for animal feed. The complete nuclear and chloroplast (cp) genome sequences of G. raimondii are already available but not mitochondria. Here, we assembled the complete mitochondrial (mt) DNA sequence of G. raimondii into a circular genome of length of 676,078 bp and performed comparative analyses with other higher plants. The genome contains 39 protein-coding genes, 6 rRNA genes, and 25 tRNA genes. We also identified four larger repeats (63.9 kb, 10.6 kb, 9.1 kb, and 2.5 kb) in this mt genome, which may be active in intramolecular recombination in the evolution of cotton. Strikingly, nearly all of the G. raimondii mt genome has been transferred to nucleus on Chr1, and the transfer event must be very recent. Phylogenetic analysis reveals that G. raimondii, as a member of Malvaceae, is much closer to another cotton (G. barbadense) than other rosids, and the clade formed by two Gossypium species is sister to Brassicales. The G. raimondii mt genome may provide a crucial foundation for evolutionary analysis, molecular biology, and cytoplasmic male sterility in cotton and other higher plants.

Assembly and comparative analysis of complete mitochondrial genome sequence of an economic plant Salix suchowensis

Willow is a widely used dioecious woody plant of Salicaceae family in China. Due to their high biomass yields, willows are promising sources for bioenergy crops. In this study, we assembled the complete mitochondrial (mt) genome sequence of S. suchowensis with the length of 644,437 bp using Roche-454 GS FLX Titanium sequencing technologies. Base composition of the S. suchowensis mt genome is A (27.43%), T (27.59%), C (22.34%), and G (22.64%), which shows a prevalent GC content with that of other angiosperms. This long circular mt genome encodes 58 unique genes (32 protein-coding genes, 23 tRNA genes and 3 rRNA genes), and 9 of the 32 protein-coding genes contain 17 introns. Through the phylogenetic analysis of 35 species based on 23 protein-coding genes, it is supported that Salix as a sister to Populus. With the detailed phylogenetic information and the identification of phylogenetic position, some ribosomal protein genes and succinate dehydrogenase genes are found usually lost during evolution. As a native shrub willow species, this worthwhile research of S. suchowensis mt genome will provide more desirable information for better understanding the genomic breeding and missing pieces of sex determination evolution in the future.

Genome-wide identification and characterization of WRKY gene family in Salix suchowensis

WRKY proteins are the zinc finger transcription factors that were first identified in plants. They can specifically interact with the W-box, which can be found in the promoter region of a large number of plant target genes, to regulate the expressions of downstream target genes. They also participate in diverse physiological and growing processes in plants. Prior to this study, a plenty of WRKY genes have been identified and characterized in herbaceous species, but there is no large-scale study of WRKY genes in willow. With the whole genome sequencing of Salix suchowensis, we have the opportunity to conduct the genome-wide research for willow WRKY gene family. In this study, we identified 85 WRKY genes in the willow genome and renamed them from SsWRKY1 to SsWRKY85 on the basis of their specific distributions on chromosomes. Due to their diverse structural features, the 85 willow WRKY genes could be further classified into three main groups (group I–III), with five subgroups (IIa–IIe) in group II. With the multiple sequence alignment and the manual search, we found three variations of the WRKYGQK heptapeptide: WRKYGRK, WKKYGQK and WRKYGKK, and four variations of the normal zinc finger motif, which might execute some new biological functions. In addition, the SsWRKY genes from the same subgroup share the similar exon–intron structures and conserved motif domains. Further studies of SsWRKY genes revealed that segmental duplication events (SDs) played a more prominent role in the expansion of SsWRKY genes. Distinct expression profiles of SsWRKY genes with RNA sequencing data revealed that diverse expression patterns among five tissues, including tender roots, young leaves, vegetative buds, non-lignified stems and barks. With the analyses of WRKY gene family in willow, it is not only beneficial to complete the functional and annotation information of WRKY genes family in woody plants, but also provide important references to investigate the expansion and evolution of this gene family in flowering plants.

Enhanced multi-weight vector projection support vector machine

Abstract Recently, we have developed an effective classifier, called Multi-weight vector projection support vector machine (MVSVM). Like traditional multisurface support vector machine Generalized-Eigenvalue-based Mulitisurface Support Vector Machine (GEPSVM), MVSVM can fast complete the computation and simultaneously handle the complex Exclusive Or (XOR) problems well. In addition, MVSVM still shows the more promising results than GEPSVM for different classification tasks. Despite the effectiveness of MVSVM, there is a serious limitation, which is that the number of the projection weight vectors for each class is limited to one. Intuitively, it is not enough to use only one projection weight vector for each class to obtain better classification. In order to address this problem, we, in this paper, develop enhanced MVSVM (EMVSVM), which is based on MVSVM. For a particular class, EMVSVM maximizes the distances from its projected average vector to the projected points from different classes to find better separability, which is different from MVSVM which maximizes the separability between classes by enforcing the maximization of the distances between the average vectors of different classes. Doing so can make EMVSVM obtain more than one discriminative weight-vector projections for each class due to that the rank of the newly-formed between-class scatter matrix is enlarged. From the statistical viewpoint, we analyze the proposed approach. Experimental results on public datasets indicate the effectiveness and efficiency of EMVSVM.

Fast orthogonal linear discriminant analysis with application to image classification

Compared to linear discriminant analysis (LDA), its orthogonalized version is a more effective statistical learning tool for dimension reduction, which devotes to better separating the data points from different classes in the lower-dimensional subspace. However, existing orthogonalized LDA techniques suffer from various drawbacks, including the requirement for expensive computing time. This paper develops an efficient orthogonal dimension reduction approach, referred to as fast orthogonal linear discriminant analysis (FOLDA), which is based on existing orthogonal linear discriminant analysis (OLDA) algorithms. However, different from previous efforts, the new approach applies the QR decomposition and the regression to solve for a new orthogonal projection vector at each iteration, leading to the by far cheaper computational cost. FOLDA achieves comparable recognition rate to existing OLDA algorithms due to the incorporation of the idea and spirit behind the latter ones. Experimental results on image databases, such as MINST, COIL20, MEPG-7 and OUTEX, show the effectiveness and efficiency of our algorithm.

Flexible orthogonal semisupervised learning for dimension reduction with image classification

In this paper, we propose a novel orthogonal manifold learning algorithm for semisupervised dimension reduction, referred to as Flexible Orthogonal Semisupervised Dimension Reduction (FODR). Our algorithm is based on the recently-developed algorithm, called Trace Ratio Based Flexible Semisupervised Discriminant Analysis (TR-FSDA). TR-FSDA introduces an orthogonality constraint and a flexible regularizer to relax such a hard linear constraint in Semisupervised Discriminant Analysis (SDA) that the low-dimensional representation is constrained to lie within the linear subspace spanned by the data, whose solution follows from solving a trace ratio problem iteratively. However, it is not guaranteed that TR-FSDA always converges. Instead of finding the orthogonal projection vectors once, our algorithm produces the orthogonal projection vectors, step by step. In each time of iterations, an orthogonal projection vector and a one-dimensional data representation are produced by solving a standard Rayleigh Quotient problem, and more importantly, the determination of a new orthogonal projection vector does not involve the knowledge of the specific statistical property for the previously-obtained orthogonal projection vectors. Therefore, it is not necessary for our FODR algorithm to guarantee the convergence. The experiments are tried out on COIL20, UMIST, ORL, YALE, MPEG-7, FERET, and Handwritten DIGIT databases, and show the effectiveness of the proposed algorithm.