Label Distribution Learning Research Articles

Unsupervised Multi-Subclass Saliency Classification for Salient Object Detection

A large number of bottom-up salient object detection algorithms formulate the problem as a classification task. For an input image, these methods tend to utilize prior cues to select some regions as training set, which can be used to learn a classifier to classify all regions into two classes: foreground/ background. However, such binary classification based approaches still suffer from accuracy problems when input image content is characterized by complex structure. To solve this problem, we propose a novel framework, namely Multi-Subclass Classification with Label Distribution Learning (MSCLDL). Specifically, prior knowledge is firstly employed to build a training set from input image, in which each sample (image region) is associated with one of two class labels (foreground/background). Previous works usually learn directly a binary classification model from training set. Different with them, we further decompose two classes into a certain number of subclasses according to image structure, each sample is thus described by one of multiple subclass labels. Based on the multi-subclass training set, we learn a label distribution model which can be treated as a classification model to predict the subclass label of each image region. Furthermore, the saliency value of each image region could be computed via exploring the relationship class and subclass labels. The MSCLDL is able to overcome the limitation of existing classification-based algorithms and produce more accurate saliency maps in some challenging scenes. Finally, a novel refinement technology is presented to further refine the saliency map obtained by MSCLDL. For fairness, we compare the proposed method and other stateof- the-art methods on four benchmark datasets, the superiority of our model is adequately demonstrated via the experimental results analysis.

Multi-label feature selection based on label distribution and neighborhood rough set

Multi-label feature selection is an indispensable technology in multi-semantic high-dimensional data preprocessing, which has been brought into focus in recent years. However, most existing methods explicitly assume that the significance of all relevant labels is the same for every instance, while ignoring the real scenarios that the significance of available labels to each instance is usually different. In this paper, we propose a novel multi-label feature selection based on label distribution and neighborhood rough set, known as LDRS. To be specific, we first construct a label enhancement method based on instance information distribution to convert the logical labels of multi-label data into label distribution, thereby capturing label significance to provide additional information for learning tasks. Then, we extend the neighborhood rough set model for label distribution learning, and discuss the related properties in detail. This extended model can effectively avoid the selection of neighborhood granularity and seamlessly apply to handle label distribution data. After that, two feature significance measures are established to realize the quality evaluation of features and the fusion of label-specific features. Finally, a novel feature selection framework is designed, which takes into account feature significance, label significance, and label-specific features, simultaneously. Experiments on both public and real-world datasets exhibit the advantages of the proposed method.

Active label distribution learning via kernel maximum mean discrepancy

Active label distribution learning via kernel maximum mean discrepancy

A Novel Probabilistic Label Enhancement Algorithm for Multi-Label Distribution Learning

We propose a novel probabilistic label enhancement algorithm, called PLEA, to solve challenging label distribution learning (LDL) for multi-label classification problems. We adopt the well-known maximum entropy model based label distribution learner. However, unlike the existing LDL algorithms based on the maximum entropy model, we propose to use manifold learning to enhance the label distribution learner. Specifically, the supervised information in the label manifold is utilized in the feature manifold space construction to improve the accuracy of feature extraction, while dramatically reducing the feature dimension. Then the robust linear regression is employed to estimate the label distributions associated with the extracted reduced-dimension features. Using the enhanced reduced-dimension features and their associated estimated label distributions in the maximum entropy model, the unknown true label distributions can be estimated more accurately, while imposing considerably lower computational complexity. We evaluate the proposed PLEA method on a wide-range artificial and high-dimensional real-world datasets. Experimental results obtained demonstrate that our proposed PLEA method has advantages in LDL accuracy and runtime performance, compared to the latest multi-label LDL approaches. The results also show that our PLEA compares favourably with the state-of-the-arts multi-label learning algorithms for classification tasks.

Multi-stage and multi-branch network with similar expressions label distribution learning for facial expression recognition

This paper proposes a novel facial expression recognition network, called Multi-Stage and Similar Expressions Label Distribution Learning Network (MSCL). Our method is based on the observations of the labels ambiguity between similar expressions in complex wild scenes, for there are inherently similar features between them that are difficult to distinguish and even manually mislabeled. The proposed network consists of three modules, namely a multi-stage multi-branch classification network (MSB), a multi-branch label distribution learning module (MLD), and a multi-branch similarity preserving module (MSP). MSB aggregates similar expression features through the first-stage prediction, MLD utilizes the aggregation results of MSB to extract the label distribution between similar expressions, and MSP utilizes a consistency relationship to minimize the differences between multiple branches. We propose an end-to-end model, where each module can be integrated with existing network modules. Furthermore, our method achieves 89.44% accuracy on the RAF-DB dataset and achieves state-of-the-art results on the AffectNet dataset with 63.25% and 66.56% accuracy on its two subsets, AffectNet-8 and AffectNet-7, respectively.

Collaborative learning network for head pose estimation

• Propose a collaborative learning framework for head pose estimation. • Learn complementary information from the landmark-based and landmark-free branches. • Design a Landmark-MLP-Mixer to obtain the head pose angles from facial landmarks. • Introduce a union loss function to optimize two complementary branches. • Conduct comparison experiments and ablation studies on various datasets. Head pose estimation is an important task in many real-world applications, such as human–computer interaction, driver monitoring, face localization and gaze estimation. In this paper, we present a novel collaborative learning framework based on Convolutional Neural Networks (CNNs) for head pose estimation. The proposed framework consists of a landmark-based branch and a landmark-free branch. The former first estimates facial landmarks and then follows the Landmark-MLP-Mixer module which models the complex nonlinear mapping relationship from facial landmarks to head pose angles. While the later adopts a label distribution learning strategy to estimate head pose. The two branches both dedicate themselves to head pose estimation task, and they collaborate with each other for mutual promotion and complementary semantic learning. Specifically, we introduce a dual-branch transfer module in the middle of the network to achieve explicit semantic interaction and introduce a multi-loss strategy that induces to implicit information interaction. We conduct extensive experiments on several popular benchmarks, including AFLW, AFLW2000 and BIWI, the results show that our method is competitive compared to other state-of-the-art methods.

Rice yield estimation using a CNN-based image-driven data assimilation framework

The accurate prediction and estimation of grain yield for smallholder farms are crucial for agricultural management . In this study, we propose an image-driven data assimilation framework that enables smallholder farmers to estimate crop yield. In the framework, a convolutional neural network was trained using label distribution learning to estimate the probability distribution of rice crop states from images. The mean and variance of the estimated probability were regarded as observed values and variances, which were then assimilated into the ORYZA2000 crop growth model to estimate the final yield. Compared with the accuracy (R 2 = 0.410, RMSE = 1.059 t ha −1 ) of yield estimation by assimilating field sampling observations, the proposed method significantly improved accuracy, with R 2 = 0.646 and RMSE = 0.679 t ha −1 . The results demonstrated the feasibility of driving a crop growth model using easily available camera images. Additionally, the probability distribution estimation method made it possible to quantify the error for each observation. The method provides a new perspective for quantifying the observation error, thereby indicating a potential research direction for data assimilation. After that, we used numerous simulation cases to determine the most valuable observation combination for final grain yield estimation, the final yield prediction ability at different development stages, and the model performance affected by number of images. Finally, a series of factors that may affect the applicability of the framework was discussed. This study establishes an intelligent data assimilation framework for smallholder farmers that can absorb crop phenotyping data from digital cameras with a low cost and simple implementation.

A label distribution manifold learning algorithm

In this paper, we propose a novel label distribution manifold learning (LDML) method for solving the multilabel distribution learning problem. First, using manifold learning, we extract the accurate and reduced-dimension features of the training data. Second, we estimate the unknown label distributions associated with the extracted reduced-dimension features based on multi-output kernel regression. Third, we use the extracted reduced-dimension features and their associated estimated label distributions to form an enhanced maximum entropy model, which enables us to accurately and efficiently estimate the unknown true label distributions for the training data. We refer to this algorithm as the LDML. We also propose to apply the tangent space alignment regression in the second stage, and the resulting algorithm is called the LDML-R. The LDML-R has better label distribution learning performance than the LDML but imposes higher complexity than the latter. We evaluate the proposed LDML and LDML-R algorithms on 15 real-world data sets with ground-truth label distributions, and the experimental results obtained show that our method has advantages in terms of learning accuracy compared to the latest multi-label distribution learning approaches. We also use another 10 real-world multi-class data sets, which do not have the ground-truth label distributions, to demonstrate the superior multilabel classification performance of our LDML-R algorithm over the existing state-of-the-art multi-label classification algorithms.

Feature selection for label distribution learning using dual-similarity based neighborhood fuzzy entropy

Label distribution learning (LDL) is a novel framework for handling label ambiguity problems and has been used widely in practice. However, dealing with high-dimensional data or data with redundant features in the LDL context is still an open problem. Existing feature selection algorithms cannot be directly applied to LDL due to the unique challenges caused by the label uncertainty nature. In this paper, we propose a novel LDL feature selection algorithm based on neighborhood rough sets. Specifically, we first introduce dual-similarity that is used to measure sample similarity in both the feature and the label spaces. Second, we invent a novel neighborhood fuzzy entropy as a feature evaluation metric, with which neighborhood rough sets can be applied to deal with LDL problems. Lastly, we complete a feature selection model that inherits the spirit of neighborhood rough sets and neighborhood fuzzy entropy. Extensive experiments have been conducted on twelve real-world LDL datasets, and the results demonstrate the superiority of our proposed model against to other six state-of-the-art algorithms.

A semi-supervised label distribution learning model with label correlations and data manifold exploration

Label distribution learning is a novel machine learning paradigm to deal with label ambiguity, which is the generalization of the traditional single-label learning and multi-label learning paradigms. Though label distribution learning has attracted a lot of attentions recently, data sets with label distributions rather than logical labels have always been scarce and most of the existing models put emphasis mainly on the supervised learning, neglecting the utilization of unlabeled samples. In this paper, we propose a semi-supervised Label Distribution Learning model with label Correlations and data Manifold exploration (sLDLCM). On one hand, sLDLCM is a semi-supervised extension of existing label distribution learning models, which can effectively make use of both labeled and unlabeled data for better capturing the underlying data properties; on the other hand, sLDLCM jointly estimates the label distributions of unlabeled samples and the other model variables. Besides, both label correlations and local data manifold are explored in sLDLCM. Extensive experiments are conducted on six real-world data sets including two facial expression, one movie rating, two bioinformatics and one visual sentiment analysis data sets. Comparative studies demonstrate that the proposed sLDLCM model achieves better performance the state-of-the-arts in terms of five evaluation metrics.

An ambiguity-aware classifier of lumbar disc degeneration

Diagnostic ambiguity is prevalent in medicine, and may lead to undesirable medical consequences. In this paper, diagnostic ambiguity for lumber disc degeneration (LDD) is quantitatively expressed as a label distribution, based on which, a classifier that can perceive and predict the diagnostic ambiguity is modeled by integrating the discal metabolomics, subjective probability quantification, and label distribution learning. An algorithm of subjective probability quantification is designed to optimally estimate the label distribution of each disc. A novel metric is proposed to evaluate the performance of the classifiers in predicting the label ranking of sample. Numerical experiments verify that the proposed classifiers are feasible to perceive and predict the ambiguous grading of LDD.

Incomplete label distribution feature selection based on neighborhood-tolerance discrimination index

Label distribution learning (LDL), focusing on the relative importance of different labels to the instance, is proposed for solving label ambiguity problem in recent years. However, for label distribution data, the annotation information may be incomplete in the real-world and complete methods cannot be directly used to process these data. In addition, with the exponential growth of data volume, data in all walks of life tend to be high-dimensional, feature selection as an efficient preprocessing technique to reduce the dimension of data. Taking the problems of the incomplete label and high-dimensional data into consideration, an incomplete label distribution feature selection method based on neighborhood-tolerance discrimination index is proposed. The neighborhood-tolerance discrimination index is utilized to explore the distinguishing ability of the feature subset, and then a novel significance metric is constructed to evaluate the importance of features, which considers the correlations between features and labels. Compared with multi-label feature selection algorithms, the proposed algorithm is designed to directly process label distribution data without discretization, which reduces information loss in the process of discretization. Compared to existing label distribution feature selection algorithms, the proposed algorithm can directly process distribution data with missing label, which avoids the interference of noisy information. Furthermore, the superiority of the proposed method over other seven state-of-the-art methods is demonstrated by conducting comprehensive experiments with eight publicly available label distribution datasets on six widely-used metrics. The experimental results show that the proposed algorithm obtains superior performance in 91.67% of cases against compared algorithms.

A Novel Ground Metric for Optimal Transport-Based Chronological Age Estimation.

Label distribution learning (LDL) is the state-of-the-art approach to dealing with a number of real-world applications, such as chronological age estimation from a face image, where there is an inherent similarity among adjacent age labels. LDL takes into account the semantic similarity by assigning a label distribution to each instance. The well-known Kullback-Leibler (KL) divergence is the widely used loss function for the LDL framework. However, the KL divergence does not fully and effectively capture the semantic similarity among age labels, thus leading to suboptimal performance. In this article, we propose a novel loss function based on the optimal transport theory for the LDL-based age estimation. A ground metric function plays an important role in the optimal transport formulation. It should be carefully determined based on the underlying geometric structure of the label space of the application in-hand. The label space in the age estimation problem has a specific geometric structure, that is, closer ages have more inherent semantic relationships. Inspired by this, we devise a novel ground metric function, which enables the loss function to increase the influence of highly correlated ages; thus exploiting the semantic similarity among ages more effectively than the existing loss functions. We then use the proposed loss function, namely, γ -Wasserstein loss, for training a deep neural network (DNN). This leads to a notoriously computationally expensive and nonconvex optimization problem. Following the standard methodology, we formulate the optimization function as a convex problem and then use an efficient iterative algorithm to update the parameters of the DNN. Extensive experiments in age estimation on different benchmark datasets validate the effectiveness of the proposed method, which consistently outperforms state-of-the-art approaches.

A Fully Deep Learning Paradigm for Pneumoconiosis Staging on Chest Radiographs.

Pneumoconiosis staging has been a very challenging task, both for certified radiologists and computer-aided detection algorithms. Although deep learning has shown proven advantages in the detection of pneumoconiosis, it remains challenging in pneumoconiosis staging due to the stage ambiguity of pneumoconiosis and noisy samples caused by misdiagnosis when they are used in training deep learning models. In this article, we propose a fully deep learning pneumoconiosis staging paradigm that comprises a segmentation procedure and a staging procedure. The segmentation procedure extracts lung fields in chest radiographs through an Asymmetric Encoder-Decoder Network (AED-Net) that can mitigate the domain shift between multiple datasets. The staging procedure classifies the lung fields into four stages through our proposed deep log-normal label distribution learning and focal staging loss. The two cascaded procedures can effectively solve the problem of model overfitting caused by stage ambiguity and noisy labels of pneumoconiosis. Besides, we collect a clinical chest radiograph dataset of pneumoconiosis from the certified radiologist's diagnostic reports. The experimental results on this novel pneumoconiosis dataset confirm that the proposed deep pneumoconiosis staging paradigm achieves an Accuracy of 90.4%, a Precision of 84.8%, a Sensitivity of 78.4%, a Specificity of 95.6%, an F1-score of 80.9% and an Area Under the Curve (AUC) of 96%. In particular, we achieve 68.4% Precision, 76.5% Sensitivity, 95% Specificity, 72.2% F1-score and 89% AUC on the early pneumoconiosis 'stage-1'.

Continuous label distribution learning

Label distribution learning (LDL) is a suitable paradigm to deal with label ambiguity through learning the correlations among different labels. Most existing label distribution learning methods consider the labels to be discrete and directly establish the mapping from features to labels. However, in many real-world applications, labels naturally form a continuous distribution, which is ignored by the existing methods. As a result, the distribution information of labels can not be accurately described and finally affects the whole learning system. The goal of this paper is to propose a novel approach which can capture the continuous distribution of different labels explicitly and effectively. Specifically, we propose Continuous Label Distribution Learning (CLDL) which describes labels as a continuous density function and learns the distribution information of the labels in the latent space. In this way, the high-order correlations among different labels can be effectively extracted and only a few parameters for describing the continuous distribution need to be learned. Extensive description degree prediction experiments on real-world datasets validate the superiority of CLDL over the existing approaches.

Design and Application of English Online Learning Platform

In order to solve the problems of the existing English online learning platforms, such as the single way of providing learning content and the unclear learning purpose, this article proposes to build a cloud computing online learning platform based on data correlation mining label distribution learning algorithm. With the help of distributed computing and virtualization technology, the software and hardware resource virtualization resource pool is integrated to build an open online learning system, which can change the defects of traditional English online learning system to the maximum extent. At the level of 10000 iterations, the average number of system requests is 2.17/s. The system runs steadily and can meet the English learning needs of student users; it can be popularized and applied in English teaching practice.

Supervised Approach to Identify Autism Spectrum Neurological Disorder via Label Distribution Learning.

Autism Spectrum Disorder (ASD) is a complicated collection of neurodevelopmental illnesses characterized by a variety of developmental defects. It is a binary classification system that cannot cope with reality. Furthermore, ASD, data label noise, high dimension, and data distribution imbalance have all hampered the existing classification algorithms. As a result, a new ASD was proposed. This strategy employs label distribution learning (LDL) to deal with label noise and uses support vector regression (SVR) to deal with sample imbalance. The experimental results show that the proposed method balances the effects of majority and minority classes on outcomes. It can effectively deal with imbalanced data in ASD diagnosis, and it can help with ASD diagnosis. This study presents a cost-sensitive approach to correct sample imbalance and uses a support vector regression (SVR)-based method to remove label noise. The label distribution learning approach overcomes high-dimensional feature classification issues by mapping samples to the feature space and then diagnosing multiclass ASD. This technique outperforms previous methods in terms of classification performance and accuracy, as well as resolving the issue of unbalanced data in ASD diagnosis.

Label distribution feature selection with feature weights fusion and local label correlations

Label distribution learning, where each sample is associated with a distribution of description degree, suffers from the curse of dimensionality like other traditional learning paradigms. Feature selection as a pre-processing technique is commonly used to reduce the dimension of data. Currently, label distribution feature selection focuses on exploiting label correlations under the assumption that all samples share the same label correlations. However, the corresponding label correlations for different groups of samples may tend to be different in real-world tasks. To tackle the issue, this paper presents a novel approach named label distribution feature selection with feature weights fusion and local label correlations. First, instances are separated into several clusters regarded as local samples. Then, a two-strategy approach is presented based on the local samples. For the first, the feature weights obtained from all local samples are fused into uniform feature significance, which can effectively improve the feature discrimination. For the second, to reflect the influence of label correlations locally, a local correlation matrix is encoded via the label space of local samples as additional features. Subsequently, the feature weights for this strategy are generated from a new feature space composed of additional and original features. Finally, a subset of optimal and relevant features is selected from the top-ranked features based on feature weights obtained from the above two strategies. Extensive experiments on fifteen benchmark datasets verify that the proposed algorithm is superior to five state-of-the-art approaches in terms of six evaluation metrics.

Selective label enhancement for multi-label classification based on three-way decisions

Multi-label classification is a challenging issue in the data science community due to the ambiguity of label semantics. Existing studies mainly focus on improving label association with logical labels, but the performance suffers from the threshold setting. Although label distribution learning gains superior discrimination, the expenditure of collecting large-scale fine-grained numerical labels is intolerable. To address the uncertainty of logical label semantics, we propose a novel model called three-way decisions with label enhancement (3WDLE). For unseen instances, we implement a trisecting-acting-outcome framework. In the trisecting stage, an uncertainty measure called global uncertain-prone degree partitions these instances into uncertain and certain regions, where the trisecting procedure is completed from label level to instance level by leveraging the distributions of pseudo-label information. In the acting stage, instances recognized as certain regions directly take the results generated by label-specific learning, whereas the remaining are reclassified by conducting selective label enhancement. The enriched knowledge generated by the label enhancement module is learnt on trustworthy instances only. In the outcome stage, we adopt five evaluation metrics to evaluate the classification performance from the perspectives of both labels and instances. In this way, three-way decisions provide a systematic methodology to deal with uncertainty in multi-label classification, which combines logical label learning with numerical label learning into a unified framework to optimize the performance of the multi-label classification model. Extensive experiments demonstrate the superiority of 3WDLE over state-of-the-art multi-label classifications with logical labels only.

Label distribution learning with noisy labels via three-way decisions

Label distribution learning (LDL) as a soft-labeling paradigm is allowed to learn single or multi-labeled information distribution. Overwhelmingly, in the open world, the distribution of labels is usually disturbed by the noise (such as the man-made induction bias, shake of hardware devices), which in turn affects the decision of downstream tasks. To address this problem, we propose a novel LDL approach by using the three-way decisions theory to clear the amplified noise in this paper. First, we evaluate the confidence of each training sample and use a three-way decisions-based method to identify the trustworthy samples and the noisy samples. Second, we apply the sample correlation between the trustworthy samples and the noisy samples to correct the noisy labels. Finally, we re-weight every sample based on the learned confidences to train the robust LDL model. Experiments show that our approach has better performance in handling noisy data compared to existing algorithms.