Search

Distance Metric Learning for Large Margin Nearest Neighbor (LMNN), as a classic distance metric learning (DML) method, has attracted much attention among researchers. However, it, like most of the existing DML methods, cannot be guaranteed to achieve independent and shared feature subspaces from multiple sources or different feature subsets, such that much statistical feature information is ignored in model learning. In this paper, we propose a novel DML model, called Multi-view DML Based on Independent and Shared Feature Subspace (MVML-ISFS), which learns multiple distance metrics to unify the information from multiple views. The proposed method finds a distance metric for each view in an independent feature space to preserve its specific property as well as a sparse representation related to the distance metrics from distinct views in a shared feature space to remain their common properties. The objective problem of MVML-ISFS is formulated based on LMNN, thus encouraging a large margin for each view that makes the distance between each of the same class pairs of samples be smaller than that between each of the different class pairs of samples. The proposed model in MVML-ISFS involves multivariate variables, which are optimized by a gradient descent strategy. The experimental results show the effectiveness of our MVML-ISFS on remote sensing, face, forest fire, and UCI datasets. [ABSTRACT FROM AUTHOR]

Accurate and real-time prediction of short-term traffic states is a crucial research topic in modern intelligent transportation systems. However, effectively modeling traffic state predictions is difficult due to the complicated characteristics of stochastic and dynamic traffic processes. In addition, collected traffic data are typically influenced by external factors (e.g. , weather, traffic jams and accidents), leading to errors and missing data. This increases the difficulty in selecting an effective method for predicting traffic conditions over time. To improve the traffic state prediction performance and alleviate the negative effect of traffic data with outliers, a novel multiclass classification least squares twin support vector machine model based on the robust L 2,p -norm (0 < p ≤ 2) distance, known as PLSTSVM, was proposed. We adjusted the PLSTSVM parameters to balance the prediction accuracy and training time. To solve the optimization problem of the PLSTSVM, an iterative algorithm was developed, which has great potential for solving other optimization problems. In addition, an integrated classification indicator system based on speed, traffic volume, occupancy rate and ample degree was used, increasing the feasibility of the traffic state analysis. To improve the learning and generalization abilities of the nonlinear PLSTSVM, we combined the polynomial kernel function and the radial basis function to construct a hybrid kernel function. The results on two real traffic datasets demonstrate that our model yields better prediction performance and robustness than other competitors, which make unsatisfactory predictions. • A robust multi-class SVM is proposed to alleviate the effect of traffic data with outliers. • An iterative algorithm is designed to solve non-smooth L 2,p -norm optimization problem. • A novel classification indicator (ample degree) is utilized to forecast the traffic state. • A hybrid kernel function is built by combining polynomial and Gaussian kernel for multi-class SVM. [ABSTRACT FROM AUTHOR]

Zero-Shot Learning (ZSL) aims to employ seen images and their related semantics to identify unseen images through knowledge transfer. Among past numerous methods, the generative methods are more prominent and achieve better results than other methods. However, we find the input for generating samples is too monotonous, there are only semantics of each class and artificially defined noise, which makes the generated visual features non-discriminative and the classifier cannot effectively distinguish them. In order to solve this problem, we propose a novel approach with cascade Generative Adversarial Network (GAN) to generate discriminative and representative features. In this method, we define a latent space where the features from different categories are orthogonal to each other and the generator for this latent space is learned with a Wasserstein GAN. In addition, in order to make up for the deficiency that the features in this latent space cannot accurately simulate the true distribution of species, we utilize another Wasserstein GAN or Cramér GAN cascaded with the previous one to generate more discriminative and representative visual features. In this way, we can not only expand the content used as input in the generation process, but also make the final generated visual features clear and separable under the influence of latent spatial orthogonality. Extensive experiments on five benchmark datasets, i.e. , AWA1, AWA2, CUB, SUN and APY, demonstrate that our proposed method can outperform most of the state-of-the-art methods on both conventional and generalized zero-shot learning settings. • Generating more representative latent and visual features to alleviate domain shift. • Features of different categories are more discriminative through orthogonal projection. • It can outperform the SOTA methods on five popular datasets. [ABSTRACT FROM AUTHOR]

Multiview Generalized Eigenvalue Proximal Support Vector Machines (MvGSVMs) is an effective multi-view classification algorithm, which effectively combines multi-view learning and classification. Then it was found that in the classification learning task, the classifier combined with multi-view learning has a better classification effect than considering only a single view. In order to utilize the multi-view learning framework more fully and accurately, we further research this. We explore the internal relationship between different views between samples to replace the method of connecting different views through distance combinations. We propose a new method named Multi-view Generalized Support Vector Machine via Mining the Inherent Relationship between Views (MRMvGSVM). At the same time, we use the L2,1-norm constraint relationship matrix as a multi-view regularization term to select the most relevant sample data from different views. It not only helps to improve the accuracy of classification but also reduces the influence of extraneous factors to a certain extent and improves the robustness of the algorithm. The effectiveness of the algorithm is proved by theory and experiments on UCI, and Face and Fire Smoke image datasets. [ABSTRACT FROM AUTHOR]

Discriminative Feature Selection (DFS) is an algorithm, proposed recently for effective feature selection by considering both joint linear discriminant analysis and row sparsity regularization. However, this method is not robust enough to protect the data from outliers, because it utilizes the squared L 2 -norm distance metric. To overcome this problem, we present in this paper, a novel discriminative feature selection algorithm, which uses the robust L 2 , 1 -norm for measuring the distances in DFS. Although the algorithm is apparently simple, it should not be considered trivial because of its non-convexity. Also, we present an analysis of the convergence, both theoretical and empirical. More importantly, we proposed an iterative algorithm to achieve optimal results. Experimental results, using various data sets, demonstrate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]