Unlabeled Ones Research Articles

Semisupervised Hyperspectral Image Classification via Discriminant Analysis and Robust Regression

In recent years, hyperspectral image classification (HSIC) has received increasing attention in a wide range of hyperspectral applications. It is still very challenging due to the following factors: 1) there are not enough labeled samples; 2) the images are easy to be polluted by outlier channels; and 3) different objects may have similar spectra. Considering these three factors, we propose a novel semisupervised HSIC method, which is constructed on discriminant analysis and robust regression (DARR). Specifically, a regression-based semisupervised technique is employed by not only exploiting the rich information in labeled samples, but also taking advantage of abundant unlabeled ones. In this way, the true data distribution can be obtained accurately. Then, we introduce a robust adaptive loss function to measure the representation loss. As a result, it can greatly relieve the side effects of outlier channels. Finally, to increase discriminating power of our approach for different objects, we utilize the pairwise constraints to incorporate the discriminant information among labeled samples. Through these constraints, the same-category samples are projected to be close to each other, while the different-category samples are as far apart as possible. The above three components can be integrated into a graph-based objective function, whose optimization is systematically provided. Extensive experiments on four data sets demonstrate that our method achieves higher quantitative results, as well as more satisfactory visual performances by comparing with state-of-the-art methods and using different parameter settings.

Online tracking based on efficient transductive learning with sample matching costs

Visual tracking has been a popular and attractive topic in computer vision for a long time. In recent decades, many challenge problems in object tracking has been effectively resolved by using learning based tracking strategies. Number of investigations carried on learning theory found that when labeled samples are limited, the learning performance can be sufficiently improved by exploiting unlabeled ones. Therefore, one of the most important issue for semi-supervised learning is how to assign the labels to the unlabeled samples, which is also the principal focus of transductive learning. Unfortunately, considering the efficiency requirement of online tracking, the optimization scheme employed by the traditional transductive learning is hard to be applied to online tracking problems because of its large computational cost during sample labeling. In this paper, we proposed an efficient transductive learning for online tracking by utilizing the correspondences among the generated unlabeled and labeled samples. Those variational correspondences are modeled by a matching costs function to achieve more efficient learning of representative separators. With a strategy of fixed budget for support vectors, the proposed learning is updated by using a weighted accumulative average of model coefficients. We evaluated the proposed tracking on benchmark database, the experiment results have demonstrated an outstanding performance via comparing with the other state-of-the-art trackers.

Multi-class semi-supervised kernel minimum squared error for face recognition

Kernel Minimum Squared Error (KMSE) has become a hot topic in machine learning and pattern recognition in the past years. However, KMSE is essentially a binary classifier and one-against-all and one-against-one strategies are usually employed to deal with multi-class problems. In this situation, KMSE needs to resolve multiple equations with the high computation complexity. Meanwhile, labeled examples are usually insufficient and unlabeled ones are abundant in many real-world applications. Therefore in this paper, we introduce a novel multi-class semi-supervised KMSE algorithm, called multi-class Laplacian regularized KMSE (McLapKMSE). Compared to KMSE and semi-supervised KMSE, we need resolve one equation at once and therefore the method has the lower complexity. The experiments on face recognition are conducted to illustrate that our algorithm can achieve the comparable performance and lower complexity in contrast to the other supervised and semi-supervised methods.

A new incremental neural network for simultaneous clustering and classification

In this paper, an Incremental Neural Network for Classification and Clustering (INNCC) is proposed. The main advantages of this neural network are the linkage between data topology preservation and classes representation by using the cluster posterior probabilities of classes. It is a constructive model without prior conditions such as a suitable number of nodes. A new neuron is inserted when new data are not represented by existing neurons. In training step, both supervised and unsupervised learning are used. The training dataset contains few samples with class labels and several unlabeled ones. The Support Vector Machines (SVM) operates in the training step to assess the INNCC classification result. The proposed approach has been tested on synthetic and real datasets. Obtained results are very promising.

Cross-domain question classification in community question answering via kernel mapping

An increasingly popular method for retrieving information is via the community question answering (CQA) systems such as Yahoo! Answers and Baidu Knows. In CQA, question classification plays an important role to find the answers. However, the labeled training examples for statistical question classifier are fairly expensive to obtain, as they require the experienced human efforts. Meanwhile, unlabeled data are readily available. This paper employs the method of domain adaptation via kernel mapping to solve this problem. In detail, the kernel approach is utilized to map the target-domain data and the source-domain data into a common space, where the question classifiers are trained under the closer conditional probabilities. The kernel mapping function is constructed by domain knowledge. Therefore, domain knowledge could be transferred from the labeled examples in the source domain to the unlabeled ones in the targeted domain. The statistical training model can be improved by using a large number of unlabeled data. Meanwhile, the Hadoop Platform is used to construct the mapping mechanism to reduce the time complexity. Map/Reduce enable kernel mapping for domain adaptation in parallel in the Hadoop Platform. Experimental results show that the accuracy of question classification could be improved by the method of kernel mapping. Furthermore, the parallel method in the Hadoop Platform could effective schedule the computing resources to reduce the running time.

Semi-supervised local ridge regression for local matching based face recognition

In this paper, a novel algorithm named Semi-supervised Local Ridge Regression (SSLRR) is proposed for local matching based face recognition. Compared with other algorithms, the proposed algorithm possesses two advantages. Firstly, SSLRR utilizes a multiple graph based semi-supervised technique to propagate the class labels of labeled samples to the unlabeled ones. Thus, the information of both labeled and unlabeled data can be employed in our algorithm to improve its performance. Secondly, unlike most local matching based face recognition algorithms which assume different sub-images from the same face are independent, an adaptive non-negative weight vector is introduced into our SSLRR to combine the Laplacian matrices obtained by different sub-images. Therefore, the latent complementary information of multiple sub-patterns from the same face image can be taken into account. Moreover, a simple yet efficient iterative update scheme is also proposed to solve our SSLRR model. Extensive experiments are performed on five standard face databases (Yale, Extended YaleB, AR, CMU PIE and LFW) to demonstrate the efficiency of the proposed algorithm. Experimental results show that SSLRR obtains better recognition performance than some other state-of-the-art approaches.

Identifying Authorities in Online Communities

Several approaches have been proposed for the problem of identifying authoritative actors in online communities. However, the majority of existing methods suffer from one or more of the following limitations: (1) There is a lack of an automatic mechanism to formally discriminate between authoritative and nonauthoritative users. In fact, a common approach to authoritative user identification is to provide a ranked list of users expecting authorities to come first. A major problem of such an approach is the question of where to stop reading the ranked list of users. How many users should be chosen as authoritative? (2) Supervised learning approaches for authoritative user identification suffer from their dependency on the training data. The problem here is that labeled samples are more difficult, expensive, and time consuming to obtain than unlabeled ones. (3) Several approaches rely on some user parameters to estimate an authority score. Detection accuracy of authoritative users can be seriously affected if incorrect values are used. In this article, we propose a parameterless mixture model-based approach that is capable of addressing the three aforementioned issues in a single framework. In our approach, we first represent each user with a feature vector composed of information related to its social behavior and activity in an online community. Next, we propose a statistical framework, based on the multivariate beta mixtures, in order to model the estimated set of feature vectors. The probability density function is therefore estimated and the beta component that corresponds to the most authoritative users is identified. The suitability of the proposed approach is illustrated on real data extracted from the Stack Exchange question-answering network and Twitter.

Labeling and qualification of endothelial progenitor cells for tracking in tissue engineering: An in vitro study.

In order to track location and distribution of endothelial cells (ECs) within scaffolds in vitro, we chose lentiPGK-TdTomato transduction of human endothelial progenitor cells (EPCs) isolated and differentiated from cord blood. Because transduction could have a functional impact on cell behavior, we checked different parameters for qualification of labeled- EPCs as well as their use for potential applications in the context of vascular and bone tissue engineering. After isolation and expansion, EPCs were classically characterized then transduced with the lentiviral vector containing the TdTomato protein gene under the control of the phosphoglycerate kinase (PGK) promoter. Conventional karyotyping, differentiation capacity, viability, proliferation assays were performed with labeled and unlabeled EPCs. Scaffolds and co-cultures were explored with labeled EPCs, in static or shear stress conditions. Our results show that cell labeling did not affect cell adhesion nor induce cell death. Cell labeling did not induce more chromosomal aberrations. Phenotypical characterization was not affected. In the context of tissue engineering applications, labeled EPCs maintained their ability to line 2D or 3D scaffolds, withstand physiological arterial shear stress, and form tubular networks in co-cultures with human osteoblast progenitor cells. It is possible to label human EPCs with TdTomato without affecting their behavior by the transduction procedure. This creates an important tool for numerous applications. Our results provide a qualification of labeled EPCs in comparison with unlabeled ones for vascular and bone tissue engineering.

Detecting positive and negative deceptive opinions using PU-learning

Nowadays a large number of opinion reviews are posted on the Web. Such reviews are a very important source of information for customers and companies. The former rely more than ever on online reviews to make their purchase decisions, and the latter to respond promptly to their clients’ expectations. Unfortunately, due to the business that is behind, there is an increasing number of deceptive opinions, that is, fictitious opinions that have been deliberately written to sound authentic, in order to deceive the consumers promoting a low quality product (positive deceptive opinions) or criticizing a potentially good quality one (negative deceptive opinions). In this paper we focus on the detection of both types of deceptive opinions, positive and negative. Due to the scarcity of examples of deceptive opinions, we propose to approach the problem of the detection of deceptive opinions employing PU-learning. PU-learning is a semi-supervised technique for building a binary classifier on the basis of positive (i.e., deceptive opinions) and unlabeled examples only. Concretely, we propose a novel method that with respect to its original version is much more conservative at the moment of selecting the negative examples (i.e., not deceptive opinions) from the unlabeled ones. The obtained results show that the proposed PU-learning method consistently outperformed the original PU-learning approach. In particular, results show an average improvement of 8.2% and 1.6% over the original approach in the detection of positive and negative deceptive opinions respectively.

A novel semi-supervised canonical correlation analysis and extensions for multi-view dimensionality reduction

Canonical correlation analysis (CCA) is an efficient method for dimensionality reduction on two-view data. However, as an unsupervised learning method, CCA cannot utilize partly given label information in multi-view semi-supervised scenarios. In this paper, we propose a novel two-view semi-supervised learning method, called semi-supervised canonical correlation analysis based on label propagation (LPbSCCA). LPbSCCA incorporates a new sparse representation based label propagation algorithm to infer label information for unlabeled data. Specifically, it firstly constructs dictionaries consisting of all labeled samples; and then obtains reconstruction coefficients of unlabeled samples using sparse representation technique; at last, by combining given labels of labeled samples, estimates label information for unlabeled ones. After that, it constructs soft label matrices of all samples and probabilistic within-class scatter matrices in each view. Finally, in order to enhance discriminative power of features, it is formulated to maximize the correlations between samples of the same class from cross views, while minimizing within-class variations in the low-dimensional feature space of each view simultaneously. Furthermore, we also extend a general model called LPbSMCCA to handle data from multiple (more than two) views. Extensive experimental results from several well-known datasets demonstrate that the proposed methods can achieve better recognition performances and robustness than existing related methods.

Online semi-supervised annotation via proxy-based local consistency propagation

Abstract In this paper, we propose a novel label propagation algorithm named Proxy-based Local Consistency Propagation (PLCP), in which the label information is first propagated from labeled examples to the unlabeled ones, and then spreads only among unlabeled ones mutually until a steady state is reached. To meet the requirements of efficiency in many real-world image annotation applications, we propose an online semi-supervised annotation framework where the new examples can be predicted and used to update the model. Specifically, we extend PLCP to work under an inductive setting and propose an incremental model updating method that can incorporate the new examples including labeled and unlabeled examples. The comprehensive experiments on MNIST, CIFAR-10 and PIE datasets show that our proposed PLCP achieves superior performance compared with the baselines, and our proposed incremental model updating method can achieve significant promotion in efficiency, with the nearly identical accuracy compared to re-training.

A General Theory of Barbs, Contexts, and Labels

Barbed bisimilarity is a widely used behavioral equivalence for interactive systems: given a set of predicates (denoted “barbs” and representing basic observations on states) and a set of contexts (representing the possible execution environments), two systems are deemed to be equivalent if they verify the same barbs whenever inserted inside any of the chosen contexts. Despite its flexibility and expressiveness, this definition of equivalence is unsatisfactory because often the quantification is over an infinite set of contexts, thus making barbed bisimilarity very hard to be verified. Should a labeled operational semantics be available, more efficient observational equivalences might be adopted. To this end, a series of techniques has been proposed to derive labeled transition systems (LTSs) from unlabeled ones, the main example being Leifer and Milner’s theory of reactive systems. The underlying intuition is that labels should be the “minimal” contexts that allow for a reduction step to be performed. However, minimality is difficult to asses, whereas the set of “intuitively” correct labels is often easily devised by the ingenuity of the researcher. This article introduces a framework that characterizes (weak) barbed bisimilarity via LTSs whose labels are (not necessarily minimal) contexts. Differently from previous proposals, our theory does not depend on the way the labeled transitions are built but instead relies on a simple set-theoretical presentation for identifying those properties such an LTS should verify to (1) capture the barbed bisimilarities of the underlying system and (2) ensure that such bisimilarities are congruences. Furthermore, we adopt suitable proof techniques to make feasible the verification of such properties. To provide a test-bed for our formalism, we instantiate it by addressing the semantics of the Mobile Ambients calculus, recasting its barbed bisimilarities via label-based behavioral equivalences.

Semi-supervised Learning for Automatic Image Annotation Based on Bayesian Framework

In this paper, we present a new method for automatic image annotation by applying semisupervised learning based on the Bayesian framework. On the one hand, we employ the semisupervised learning, i.e., transductive support vector machine (TSVM) to enhance the quality of training image data, which is a promising way to find out the underlying relevant data from the unlabeled ones. On the other hand, a simple yet very efficient Bayesian model is built to implement image annotation by the maximum a posteriori (MAP) criterion. The novelty of our method mainly lies in two aspects: exploiting TSVM to improve the quality of training image dataset and utilizing the Bayesian model to predict the candidate annotations for the unseen images. Experimental results on the standard Corel dataset demonstrate that the proposed method is superior or highly competitive to several state-of-the-art approaches.

Semi-Supervised Transductive Hot Spot Predictor Working on Multiple Assumptions

Protein-protein interactions are critically dependent on just a few residues (“hot spots”) at the interfaces. Hot spots make a dominant contribution to the binding free energy and if mutated they can disrupt the interaction. As mutagenesis studies require significant experimental efforts, there exists a need for accurate and reliable computational hot spot prediction methods. Compared to the supervised hot spot prediction algorithms, the semi-supervised prediction methods can take into consideration both the labeled and unlabeled residues in the dataset during the prediction procedure. The transductive support vector machine has been utilized for this task and demonstrated a better prediction performance. To the best of our knowledge, however, none of the transductive semi-supervised algorithms takes all the three semisupervised assumptions, i.e., smoothness, cluster and manifold assumptions, together into account during learning. In this paper, we propose a novel semi-supervised method for hot spot residue prediction, by considering all the three semisupervised assumptions using nonlinear models. Our algorithm, IterPropMCS, works in an iterative manner. In each iteration, the algorithm first propagates the labels of the labeled residues to the unlabeled ones, along the shortest path between them on a graph, assuming that they lie on a nonlinear manifold. Then it selects the most confident residues as the labeled ones for the next iteration, according to the cluster and smoothness criteria, which is implemented by a nonlinear density estimator. Experiments on a benchmark dataset, using protein structure-based features, demonstrate that our approach is effective in predicting hot spots and compares favorably to other available methods. The results also show that our method outperforms the state-of-the-art transductive learning methods.

Active learning with multi-criteria decision making systems

In active learning, the learner is required to measure the importance of unlabeled samples in a large dataset and select the best one iteratively. This sample selection process could be treated as a decision making problem, which evaluates, ranks, and makes choices from a finite set of alternatives. In many decision making problems, it usually applied multiple criteria since the performance is better than using a single criterion. Motivated by these facts, an active learning model based on multi-criteria decision making (MCMD) is proposed in this paper. After the investigation between any two unlabeled samples, a preference preorder is determined for each criterion. The dominated index and the dominating index are then defined and calculated to evaluate the informativeness of unlabeled samples, which provide an effective metric measure for sample selection. On the other hand, under multiple-instance learning (MIL) environment, the instances/samples are grouped into bags, a bag is negative only if all of its instances are negative, and is positive otherwise. Multiple-instance active learning (MIAL) aims to select and label the most informative bags from numerous unlabeled ones, and learn a MIL classifier for accurately predicting unseen bags by requesting as few labels as possible. It adopts a MIL algorithm as the base classifier, and follows an active learning procedure. In order to achieve a balance between learning efficiency and generalization capability, the proposed active learning model is restricted to a specific algorithm under MIL environment. Experimental results demonstrate the effectiveness of the proposed method.

Learning and inspecting classification rules from longitudinal epidemiological data to identify predictive features on hepatic steatosis

Personalized medicine requires the analysis of epidemiological data for the identification of subgroups sharing some risk factors and exhibiting dedicated outcome risks. We investigate the potential of data mining methods for the analysis of subgroups of cohort participants on hepatic steatosis. We propose a workflow for data preparation and mining on epidemiological data and we present InteractiveRuleMiner, an interactive tool for the inspection of rules in each subpopulation, including functionalities for the juxtaposition of labeled individuals and unlabeled ones. We report on our insights on specific subpopulations that have been discovered in a data-driven rather than hypothesis-driven way.

2013年生态学术年会论文——遮荫对米槠和杉木原位排放甲烷的影响

Although there were controversies on whether terrestrial plants emit methane( CH4) under aerobic conditions, more and more researches confirmed that plants themselves did emit CH4.,and sunlight was one of the important factors which effected the CH4emissions from plants. In this study,we used stable carbon isotope-labelling technique combining with potted experiments to discussed the source of methane emitted in situ from Castanopsis carlesii and Cunninghamia lanceolata and their responses to sunlight and shadowing. The results showed that there were significant differences in the stable carbon isotopic value of labeled and unlabeled leaves between Castanopsis carlesii and Cunninghamia lanceolata,the stable carbon isotopic value( δ13C) of the labeled Castanopsis carlesii and Cunninghamia lanceolata leaves was 128.9.and 71.1 times higher than that of the unlabeled ones. Similarly,there were significant differences in the stable carbon isotopic value of methane from the labeled branch leaves of Castanopsis carlesii and Cunninghamia lanceolata and that from the unlabeled ones. Been sealed 2 hours,the stable carbon isotopic value of methane( δ13CH4) isotopic signatures of methane, which emitted from Castanopsis carlesii and Cunninghamia lanceolata were 21. 1 and 28. 2 times higher than that of the unlabeled ones. These results indicated that the most detected CH4emission during our experiments were from plants themselves. Shadowing and natural lighting experiments showed that the CH4emission rate of Cunninghamia lanceolata was29% higher than Castanopsis carlesii under sunlight,the mean CH4emission rates of them were 6.3 and 4.88 ng CH4·g-1DW ·h-1,respectively. The fluctuation rule of CH4emission rates are similar between the two species under sunlight,both of them were reduced at first,and then increased,finally,they were reduced again. However,the rule was different under shadowing,the mean CH4emission rate of Cunninghamia lanceolata was 23% lower than under sunlight,especially after 15 days,the CH4emission rate reduced significantly,until June 30th,it was only 15.1% of that detected under sunlight. The mean CH4emission rate of Castanopsis carlesii were not significantly different under sunlight and shadowing,they were 6.3 and 5.67ng CH4·g-1DW·h-1,respectively; but after 5 days,the CH4emission rate was 71% higher than under sunlight,it reached up to 10.45 ng CH4·g-1DW·h-1,then it dropped sharply,and obviously lower than under sunlight,until June 30th, it was only 43.1% of that detected under sunlight. These results suggested that sunlight promoted methane emission rate from plants.

Self-training-based face recognition using semi-supervised linear discriminant analysis and affinity propagation

Face recognition is one of the most important applications of machine learning and computer vision. The traditional supervised learning methods require a large amount of labeled face images to achieve good performance. In practice, however, labeled images are usually scarce while unlabeled ones may be abundant. In this paper, we introduce a semi-supervised face recognition method, in which semi-supervised linear discriminant analysis (SDA) and affinity propagation (AP) are integrated into a self-training framework. In particular, SDA is employed to compute the face subspace using both labeled and unlabeled images, and AP is used to identify the exemplars of different face classes in the subspace. The unlabeled data can then be classified according to the exemplars and the newly labeled data with the highest confidence are added to the labeled data, and the whole procedure iterates until convergence. A series of experiments on four face datasets are carried out to evaluate the performance of our algorithm. Experimental results illustrate that our algorithm outperforms the other unsupervised, semi-supervised, and supervised methods.

Interactive shape co-segmentation via label propagation

In this paper, we present an interactive approach for shape co-segmentation via label propagation. Our intuitive approach is able to produce error-free results and is very effective at handling out-of-sample data. Specifically, we start by over-segmenting a set of shapes into primitive patches. Then, we allow the users to assign labels to some patches and propagate the label information from these patches to the unlabeled ones. We iterate the last two steps until the error-free consistent segmentations are obtained. Additionally, we provide an inductive extension of our framework, which effectively addresses the out-of-sample data. The experimental results demonstrate the effectiveness of our approach.

Random walks in directed hypergraphs and application to semi-supervised image segmentation

In this paper, we introduce for the first time the notion of directed hypergraphs in image processing and particularly image segmentation. We give a formulation of a random walk in a directed hypergraph that serves as a basis to a semi-supervised image segmentation procedure that is configured as a machine learning problem, where a few sample pixels are used to estimate the labels of the unlabeled ones. A directed hypergraph model is proposed to represent the image content, and the directed random walk formulation allows to compute a transition matrix that can be exploited in a simple iterative semi-supervised segmentation process. Experiments over the Microsoft GrabCut dataset have achieved results that demonstrated the relevance of introducing directionality in hypergraphs for computer vision problems.