In this article, we propose a semi-supervised non-negative matrix factorization (NMF) model by means of elegantly modeling the label information. The proposed is capable generating discriminable low-dimensional representations to improve clustering performance. Specifically, pair complementary regularizers, i.e., similarity and dissimilarity incorporated into conventional NMF guide factorization. And, they impose restrictions on both data samples with labels as well small number unlabeled...

The combination of the traditional convolutional network (i.e., an auto-encoder) and graph has attracted much attention in clustering, which auto-encoder extracts node attribute feature captures topological feature. However, existing works (i) lack a flexible mechanism to adaptively fuse those two kinds features for learning discriminative representation (ii) overlook multi-scale information embedded at different layers subsequent cluster assignment, leading inferior clustering results. To...

This paper explores the problem of multi-view spectral clustering (MVSC) based on tensor low-rank modeling. Unlike existing methods that all adopt an off-the-shelf norm without considering special characteristics in MVSC, we design a novel structured tailored to MVSC. Specifically, explicitly impose symmetric constraint and sparse frontal horizontal slices characterize intra-view inter-view relationships, respectively. Moreover, two constraints could be jointly optimized achieve mutual...

Ensemble clustering integrates a set of base results to generate stronger one. Existing methods usually rely on co-association (CA) matrix that measures how many times two samples are grouped into the same cluster according clusterings achieve ensemble clustering. However, when constructed CA is low quality, performance will degrade. In this article, we propose simple, yet effective self-enhancement framework can improve better performance. Specifically, first extract high-confidence (HC)...

In this paper, a joint machine learning and game theory modeling (MLGT) framework is proposed for inter frame coding tree unit (CTU) level bit allocation rate control (RC) optimization in high efficiency video (HEVC). First, support vector machine-based multi-classification scheme to improve the prediction accuracy of CTU-level rate-distortion (R-D) model. The legacy "chicken-and-egg" dilemma be overcome by learning-based R-D Second, mixed model-based cooperative bargaining optimization,...

As a variant of nonnegative matrix factorization (NMF), symmetric NMF (SNMF) has shown to be effective for capturing the cluster structure embedded in graph representation. In contrast existing SNMF-based clustering methods that empirically construct similarity and rigidly introduce supervisory information assignment matrix, this paper, we propose novel semisupervised method, namely, pairwise constraint propagation-induced SNMF (PCPSNMF). By formulating single-constrained optimization...

As a variant of non-negative matrix factorization (NMF), symmetric NMF (SymNMF) can generate the clustering result without additional post-processing, by decomposing similarity into product indicator and its transpose. However, in traditional SymNMF methods is usually predefined, resulting limited performance. Considering that quality graph crucial to final performance, we propose new semisupervised model, which able simultaneously learn with supervisory information results, such mutual...

Deep self-expressiveness-based subspace clustering methods have demonstrated effectiveness. However, existing works only consider the attribute information to conduct self-expressiveness, which may limit performance. In this paper, we propose a novel adaptive and structure network (AASSC-Net) simultaneously in an graph fusion manner. Specifically, first exploit auto-encoder represent input data samples with latent features for construction of matrix. We also construct mixed signed symmetric...

Exploiting label correlations is important to multi-label classification. Previous methods capture the high-order mainly by transforming matrix a latent space with low-rank factorization. However, generally full-rank or approximate matrix, making factorization inappropriate. Besides, in space, will become implicit. To this end, we propose simple yet effective method depict explicitly, and at same time maintain high-rank of matrix. Moreover, estimate infer model parameters simultaneously via...

Multi-label learning (MLL) has gained attention for its ability to represent real-world data. Label Distribution Learning (LDL), an extension of MLL from label distributions, faces challenges in collecting accurate distributions. To address the issue biased annotations, based on low-rank assumption, existing works recover true distributions observations by exploring correlations. However, recent evidence shows that distribution tends be full-rank, and naive apply approximation observation...

Despite the remarkable success of deep neural networks (DNNs), security threat adversarial attacks poses a significant challenge to reliability DNNs. By introducing randomness into different parts DNNs, stochastic methods can enable model learn some uncertainty, thereby improving robustness efficiently. In this paper, we theoretically discover universal phenomenon that will shift distributions feature statistics. Motivated by theoretical finding, propose enhancement module called Feature...

By transforming a multi-objective optimization problem into number of single-objective problems and optimizing them simultaneously, decomposition-based evolutionary algorithms have attracted much attention in the field optimization. In algorithms, population diversity is maintained using set predefined weight vectors, which are often evenly sampled on unit simplex. However, when Pareto front not hyperplane but more complex, distribution final solution will be that uniform. this paper, we...

Existing deep embedding clustering methods fail to sufficiently utilize the available off-the-shelf information from feature embeddings and cluster assignments, limiting their performance. To this end, we propose a novel method, namely attention-guided graph with dual self-supervision (DAGC). Specifically, DAGC first utilizes heterogeneity-wise fusion module adaptively integrate features of auto-encoder convolutional network in each layer then uses scale-wise dynamically concatenate...

In this letter, we propose a novel semi-supervised subspace clustering method, which is able to simultaneously augment the initial supervisory information and construct discriminative affinity matrix. By representing limited amount of as pairwise constraint matrix, observe that ideal matrix for shares same low-rank structure Thus, stack two matrices into 3-D tensor, where global imposed promote construction constraints synchronously. Besides, use local geometry input samples complement prior...

Spectral clustering (SC) is one of the most widely used methods. In this letter, we extend traditional SC with a semi-supervised manner. Specifically, guidance small amount supervisory information, build matrix anti-block-diagonal appearance, which further utilized to regularize product low-dimensional embedding and its transpose. Technically, formulate proposed model as constrained optimization problem. Then, relax it convex problem, can be efficiently solved global convergence guaranteed...

Nonnegative matrix factorization (NMF) is a well-known paradigm for data representation. Traditional NMF-based classification methods first perform NMF or one of its variants on input samples to obtain their low-dimensional representations, which are successively classified by means typical classifier [e.g., k -nearest neighbors (KNN) and support vector machine (SVM)]. Such stepwise manner may overlook the dependency between two processes, resulting in compromise accuracy. In this paper, we...

Deep subspace clustering networks have attracted much attention in clustering, which an auto-encoder non-linearly maps the input data into a latent space, and fully connected layer named self-expressiveness module is introduced to learn affinity matrix via typical regularization term (e.g., sparse or low-rank). However, adopted terms ignore connectivity within each subspace, limiting their performance. In addition, framework suffers from coupling issue between module, making network training...

In this paper, we propose a novel classification scheme for the remotely sensed hyperspectral image (HSI), namely SP-DLRR, by comprehensively exploring its unique characteristics, including local spatial information and low-rankness. SP-DLRR is mainly composed of two modules, i.e., classification-guided superpixel segmentation discriminative low-rank representation, which are iteratively conducted. Specifically, utilizing incorporating predictions from typical classifier, first module...

Today, the development of unmanned aerial vehicles (UAVs) has attracted significant attention in both civil and military fields due to their flight flexibility complex dangerous environments. However, energy constraints, UAVs can only finish a few tasks limited time. The problem finding best path while balancing task completion time coverage rate needs be resolved urgently. Therefore, this paper proposes UAV algorithm base on greedy strategy ant colony optimization. Firstly, introduces...

Positive label is often used as the supervisory information in learning scenario, which refers to category that a sample assigned to. However, another side lying labels, describes categories exclusive of, have been largely ignored. In this paper, we propose nonnegative matrix factorization (NMF) based classification method leveraging both positive and negative information, termed label-driven NMF (PNLD-NMF). The proposed scheme concurrently accomplishes data representation joint manner....

In this article, we propose a novel model for constrained clustering, namely, the dissimilarity propagation-guided graph-Laplacian principal component analysis (DP-GLPCA). By fully utilizing limited number of weakly supervisory information in form pairwise constraints, proposed DP-GLPCA is capable capturing both local and global structures input samples to exploit their characteristics excellent clustering. More specifically, first formulate convex semisupervised low-dimensional embedding by...