Search

EEG signal classification is a key technology for EEG signal processing and identification systems. Dictionary learning has shown excellent performance due to its sparse representation and learning capability. Usually dictionary learning requires sufficient labeled EEG signals to build classification models, and assumes that the data distribution of training and test signals are the same. However, in new EEG signal domain, often only a small amount of signals are labeled, and more are not labeled. At the same time, data dynamicity, confounding factors and strong interclass similarity also seriously disrupt the performance of EEG signal classifier. To this end, a geometric preserving transfer discriminative dictionary learning method called GPTDDL is developed for cross-domain EEG signal classification. Through projected signals of different domains to the common subspace, a shared discriminative dictionary is obtained, which explores the geometric structure information by graph Laplacian regularization and discriminative information by principal component analysis regularization. Benefiting from the discriminative information transferred from source domain, the discriminability of the learned sparse coding of target domain is strengthened. GPTDDL integrates this idea into the framework of LC-KSVD, and learns the subspace and dictionary learning parameters in an iterative strategy. The experimental results on Bonn EEG signal dataset demonstrate the validity of the GPTDDL method.

Facial expression recognition has drawn increasing attention because of its great potential in human behavior analysis. Traditional recognition models usually assume that the training set is sufficient, and the training and testing data sets have the same distribution. However, these two factors are not satisfied in some cases. In this study, a transfer model collaborating metric learning and dictionary learning called TMMLDL is proposed to address the transfer facial expression recognition problem. To reduce the impact of cross-domain distribution variation, the information of global structure and pairwise constraints are utilized among training images in different domains. In particular, a discriminative metric space is learned into dictionary learning procedure such that the dictionary items can well present the discriminative information of different facial expression classes in the metric subspace. The proposed model tunes dictionary and metric space in an alternative optimization algorithm, which is guaranteed to obtain the optimal model parameters simultaneously. The experimental results on nine cross-domain facial expression classification tasks show that the proposed model achieves the satisfactory recognition performance.

Traditional clustering methods are not very effective when dealing with high-dimensional and huge datasets. Even if there are some traditional dimensionality reduction methods such as Principal components analysis (PCA), Linear discriminant analysis (LDA) and T-distributed stochastic neighbor embedding (T-SNE), they still can not significantly improve the effect of the clustering algorithm in this scenario. Recent studies have combined Non-linear dimensionality reduction achieved by deep neural networks with hard-partition clustering, and have achieved reliability results, but these methods can not update the parameters of dimensionality reduction and clustering at the same time. We found that soft-partition clustering can be well combined with deep embedding, and the membership of Fuzzy c-means (FCM) can solve the problem that gradient descent can not be implemented because the assignment process of the hard-partition clustering algorithm is discrete, so that the algorithm can update the parameters of deep neural network (DNN) and cluster centroids at the same time. We build an continuous objective function that combine the soft-partition clustering with deep embedding, so that the learning representations can be cluster-friendly. The experimental results show that our proposed method of simultaneously optimizing the parameters of deep dimensionality reduction and clustering is better than the method with separate optimization.

EEG signal classification play an important role in recognition of epilepsy. Recently, dictionary learning algorithms have shown the effectiveness in this field. When designing dictionaries, due to highly non-stationary of EEG signals, and collecting signals existing in different stimulus and drug modes, training and testing scenarios may be different. Thus, the performance of classical dictionary learning algorithms is unsatisfactory. In this paper, a transfer discriminative dictionary learning with label consistency (called TDDLLC) algorithm is proposed for EEG signal classification. Since each EEG signal can be represented as a linear combination of dictionary atoms, and some atoms are dataset independent, two dictionaries are learned simultaneously in source domain (SD) and target domain (TD) respectively where the discrepancy between two dictionaries is minimized. Meanwhile, utilizing the label information of samples in SD and a small number of labeled samples in TD, these dictionaries are learned with the aim of achieving discriminative abilities. To avoid the NP-hard problem, $$\ell_{1}$$ -norm regularization term is used in TDDLLC, and objective function is solved by block-coordinate descent method. Extensive experiments have been performed on Bonn dataset and show the validity of the TDDLLC algorithm.

Online support vector machine (SVM) is an effective learning method in real-time network traffic classification tasks. However, due to its geometric characteristics, the traditional online SVMs are sensitive to noise and class imbalance. In this paper, a scalable kernel convex hull online SVM called SKCHO-SVM is proposed to solve this problem. SKCHO-SVM involves two stages: (1) offline leaning stage, in which the noise points are deleted and initial pin-SVM classifier is built; (2) online updating stage, in which the classifier is updated with newly arrived data points, while carrying out the classification task. The noise deleting strategy and pinball loss function ensure SKCHO-SVM insensitive to noise data flows. Based on the scalable kernel convex hull, a small amount of convex hull vertices are dynamically selected as the training data points in each class, and the obtained scalable kernel convex hull can relieve class imbalance. Theoretical analysis and numerical experiments show that SKCHO-SVM has the distinctive ability of training time and classification performance.

Recognizing noncoding ribonucleic acid (ncRNA) data is helpful in realizing the regulation of tumor formation and certain aspects of life mechanisms, such as growth, differentiation, development, and immunity. However, the scale of ncRNA data is usually very large. Using machine learning (ML) methods to automatically analyze these data can obtain more precise results than manually analyzing these data, but the traditional ML algorithms can process only small-scale training data. To solve this problem, a novel multitask cross-learning 0-order Takagi–Sugeno–Kang fuzzy classifier (MT-CL-0-TSK-FC) is proposed that uses a multitask cross-learning mechanism to solve the large-scale learning problem of ncRNA data. In addition, the proposed MT-CL-0-TSK-FC method naturally inherits the interpretability of traditional fuzzy systems and eventually generates an interpretable rulesbased database to recognize the ncRNA data. The experimental results indicate that the proposed MT-CL-0TSK-FC method has a faster running time and better classification accuracy than traditional ML methods.

Image based face verification has attracted extension attention in the fields of pattern recognition and intelligent vision. With difference in age, cross-age face verification from facial images remains a challenging work because of a large number of facial variations caused by shape, skin color and wrinkles and so on. This study proposes a multi-task deep metric learning with boundary discriminative information method called MDML-BDI. It learns a distance metric by exploring discriminative information among the interclass neighborhood samples, such that the distances between intraclass samples are as small as possible and that between interclass neighborhood samples are as far as possible. MDML-BDI learns hierarchical nonlinear transformations by integrating metric learning into the framework of multi-task deep neural network, such that a common shared layer shares the common transformation by multiple tasks, and the other independent layers learn individual task-special transformation for each task. Experimental results on FG-NET, CACD-VS and CALFW datasets show that MDML-BDI achieves satisfactory performance in terms of accuracy and receiver operating characteristic (ROC) curve.

Support vector machine (SVM) achieves successful classification performance with the application in non-coding RNA (ncRNA) data. With the rapid increase of the species and sizes of ncRNA sequences, several fast SVM methods based on data distribution and contour information have been developed to reduce their time complexity. However, they are sensitive to both noise and class imbalance problems. In this paper, a fast and robust SVM with anti-noise convex hull for large-scale ncRNA data classification (called FRSVM-ANCH) is proposed. FRSVM-ANCH discards the outliers in the feature space and obtains the convex hull of different classes. Then, the convex hull as the training data, along with its weight is used to train the SVM. Due to less sensitive to noise, pinball loss is adopted in SVM classifier. Theoretical analysis and experimental results verify the advantages of FRSVM-ANCH in classification performance and training time on large scale noisy and imbalanced ncRNA datasets.

The rule-based fuzzy systems have successfully applied for numerous medical data classification problems. However, structuring the concise and interpretable fuzzy rules with good classification performance is still a big challenge. To address this issue, a novel feature selection and rule generation integrated learning for Takagi-Sugeno-Kang fuzzy system (called FSRG-IL-TSK) in this paper. FSRG-IL-TSK represents feature selection, structure identification and parameter learning into a Bayesian model, and uses the sequential importance resampling (SIR) algorithm to obtain the optimal parameters simultaneously, including the optimal features for each fuzzy rule, number of rules, and antecedent/consequent parameter of rules. Due to an integrated learning mechanism, it can select a small set of useful features and obtain a small number of rules. The effectiveness and advantages of FSRG-IL-TSK are validated experimentally on real-world medical data classification tasks.

Background: The brain magnetic resonance imaging (MRI) image segmentation method mainly refers to the division of brain tissue, which can be divided into tissue parts such as white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF). The segmentation results can provide a basis for medical image registration, 3D reconstruction, and visualization. Generally, MRI images have defects such as partial volume effects, uneven grayscale, and noise. Therefore, in practical applications, the segmentation of brain MRI images has difficulty obtaining high accuracy.Materials and Methods: The fuzzy clustering algorithm establishes the expression of the uncertainty of the sample category and can describe the ambiguity brought by the partial volume effect to the brain MRI image, so it is very suitable for brain MRI image segmentation (B-MRI-IS). The classic fuzzy c-means (FCM) algorithm is extremely sensitive to noise and offset fields. If the algorithm is used directly to segment the brain MRI image, the ideal segmentation result cannot be obtained. Accordingly, considering the defects of MRI medical images, this study uses an improved multiview FCM clustering algorithm (IMV-FCM) to improve the algorithm’s segmentation accuracy of brain images. IMV-FCM uses a view weight adaptive learning mechanism so that each view obtains the optimal weight according to its cluster contribution. The final division result is obtained through the view ensemble method. Under the view weight adaptive learning mechanism, the coordination between various views is more flexible, and each view can be adaptively learned to achieve better clustering effects.Results: The segmentation results of a large number of brain MRI images show that IMV-FCM has better segmentation performance and can accurately segment brain tissue. Compared with several related clustering algorithms, the IMV-FCM algorithm has better adaptability and better clustering performance.

Metric learning is a class of efficient algorithms for EEG signal classification problem. Usually, metric learning method deals with EEG signals in the single view space. To exploit the diversity and complementariness of different feature representations, a new auto-weighted multi-view discriminative metric learning method with Fisher discriminative and global structure constraints for epilepsy EEG signal classification called AMDML is proposed to promote the performance of EEG signal classification. On the one hand, AMDML exploits the multiple features of different views in the scheme of the multi-view feature representation. On the other hand, considering both the Fisher discriminative constraint and global structure constraint, AMDML learns the discriminative metric space, in which the intraclass EEG signals are compact and the interclass EEG signals are separable as much as possible. For better adjusting the weights of constraints and views, instead of manually adjusting, a closed form solution is proposed, which obtain the best values when achieving the optimal model. Experimental results on Bonn EEG dataset show AMDML achieves the satisfactory results.

Epilepsy is an abnormal function disease of movement, consciousness, and nerve caused by abnormal discharge of brain neurons in the brain. EEG is currently a very important tool in the process of epilepsy research. In this paper, a novel noise-insensitive Takagi-Sugeno-Kang (TSK) fuzzy system based on interclass competitive learning is proposed for EEG signal recognition. First, a possibilistic clustering in Bayesian framework with interclass competitive learning called PCB-ICL is presented to determine antecedent parameters of fuzzy rules. Inherited by the possibilistic c-means clustering, PCB-ICL is noise insensitive. PCB-ICL learns cluster centers of different classes in a competitive relationship. The obtained clustering centers are attracted by the samples of the same class and also excluded by the samples of other classes and pushed away from the heterogeneous data. PCB-ICL uses the Metropolis-Hastings method to obtain the optimal clustering results in an alternating iterative strategy. Thus, the learned antecedent parameters have high interpretability. To further promote the noise insensitivity of rules, the asymmetric expectile term and Ho-Kashyap procedure are adopted to learn the consequent parameters of rules. Based on the above ideas, a TSK fuzzy system is proposed and is called PCB-ICL-TSK. Comprehensive experiments on real-world EEG data reveal that the proposed fuzzy system achieves the robust and effective performance for EEG signal recognition.

Classification of electroencephalogram (EEG) signal data plays a vital role in epilepsy detection. Recently sparse representation-based classification (SRC) methods have achieved the good performance in EEG signal automatic detection, by which the EEG signals are sparsely represented using a few active coefficients in the dictionary and classified according to the reconstruction criteria. However, most of SRC learn a linear dictionary for encoding, and cannot extract enough information and nonlinear relationship of data for classification. To solve this problem, a hierarchical discriminative sparse representation classification model (called HD-SRC) for EEG signal detection is proposed. Based on the framework of neural network, HD-SRC learns the hierarchical nonlinear transformation and maps the signal data into the nonlinear transformed space. Through incorporating this idea into label consistent K singular value decomposition (LC-KSVD) at the top layer of neural network, HD-SRC seeks discriminative representation together with dictionary, while minimizing errors of classification, reconstruction and discriminative sparse-code for pattern classification. By learning the hierarchical feature mapping and discriminative dictionary simultaneously, more discriminative information of data can be exploited. In the experiment the proposed model is evaluated on the Bonn EEG database, and the results show it obtains satisfactory classification performance in multiple EEG signal detection tasks.

Conventional classification models for epileptic EEG signal recognition need sufficient labeled samples as training dataset. In addition, when training and testing EEG signal samples are collected from different distributions, for example, due to differences in patient groups or acquisition devices, such methods generally cannot perform well. In this paper, a cross-domain classification model with knowledge utilization maximization called CDC-KUM is presented, which takes advantage of the data global structure provided by the labeled samples in the related domain and unlabeled samples in the current domain. Through mapping the data into kernel space, the pairwise constraint regularization term is combined together the predictive differences of the labeled data in the source domain. Meanwhile, the soft clustering regularization term using quadratic weights and Gini-Simpson diversity is applied to exploit the distribution information of unlabeled data in the target domain. Experimental results show that CDC-KUM model outperformed several traditional non-transfer and transfer classification methods for recognition of epileptic EEG signals.

A novel discriminative deep transfer learning method called DDTML is proposed for Cross-scenario Person Reidentification( Re-ID). Using a deep neural network, DDTML learns a set of hierarchical nonlinear transformations for Cross-scenario Person Re-identification by transferring discriminative knowledge from the source domain to the target domain. Meanwhile, taking account of the inherent characteristics of Re-ID data sets, in order to reduce the distribution divergence between the source data and the target data, DDTML minimizes a new maximum mean discrepancy based on Class Distribution called MMDCD at the top layer of the network. Experimental results on widely used Re-identification datasets show the effectiveness of the proposed classifiers.

This paper focus on distributions of each class and proposes a novel person re-identification method using K-L divergence in the metric learning stage. The metric learning is not directly based on images or features, but directly based on distributions. The key idea of this paper is to assume that each person is a distribution and each image of a person is an instance of the distribution. Recognizing a probe becomes a task to determine which distribution the probe belongs to. In further, it assumes that the features of a person follow a multivariate Gaussian distribution and different people's distributions are different only with means of features but are same in their covariance matrices. The learning process is to find a global optimal covariance matrix among features for all of the distributions. A probe is then classified by comparing the K-L divergence with each class (distribution). The major contribution of this paper lies on the idea of distribution based metric learning methods, which is significantly different from most of the existing methods. Since the learning is among distributions, not images, the proposed model significantly reduced the computational cost and computational complexity and is much faster than traditional methods while the recognition rate is still quite competitive.

A novel support vector machine with manifold regularization and partially labeling privacy protection, termed as SVM-MR&PLPP, is proposed for semi-supervised learning (SSL) scenarios where only few labeled data and the class proportion of unlabeled data, due to privacy protection concerns, are available. It integrates manifold regularization and privacy protection regularization into the Laplacian support vector machine (LapSVM) to improve the classification accuracy. Privacy protection here refers to use only the class proportion of data. In order to circumvent the high computational burden of the matrix inversion operation involved in SVM-MR&PLPP, its scalable version called SSVM-MR&PLPP is further developed by introducing intermediate decision variables into the original regularization framework so that the computational burden of the corresponding transformed kernel in SSVM-MR&PLPP can be greatly reduced, making it highly scalable to large datasets. The experimental results on numerous datasets show the effectiveness of the proposed classifiers.

Existing projection-based person re-identification methods usually suffer from long time training, high dimension of projection matrix, and low matching rate. In addition, the intra-class instances may be much less than the inter-class instances when a training data set is built. To solve these problems, a novel relative distance metric leaning based on clustering centralization and projection vectors learning is proposed. When constructing training data set, the images of a same target person are clustering centralized with fuzzy c-means). The training data sets are built by these clusters in order to alleviate the imbalanced data problem of the training data sets. In addition, during learning projection matrix, the resulted projection vectors can be approximately orthogonal by using an iteration strategy and a conjugate gradient projection vector learning method to update training data sets. Experimental results show that the proposed algorithm has higher efficiency. The matching rate can be significantly improved, and the time of training is much shorter than most of existing algorithms of person re-identification.

Distributed denial of service (DDoS) attacks are one of the major threats to the current Internet, and application-layer DDoS attacks utilizing legitimate HTTP requests to overwhelm victim resources are more undetectable. Consequently, neither intrusion detection systems (IDS) nor victim server can detect malicious packets. In this paper, a novel approach to detect application-layer DDoS attack is proposed based on entropy of HTTP GET requests per source IP address (HRPI). By approximating the adaptive autoregressive (AAR) model, the HRPI time series is transformed into a multidimensional vector series. Then, a trained support vector machine (SVM) classifier is applied to identify the attacks. The experiments with several databases are performed and results show that this approach can detect application-layer DDoS attacks effectively.

Conventional, soft-partition clustering approaches, such as fuzzy c-means (FCM), maximum entropy clustering (MEC) and fuzzy clustering by quadratic regularization (FC-QR), are usually incompetent in those situations where the data are quite insufficient or much polluted by underlying noise or outliers. In order to address this challenge, the quadratic weights and Gini-Simpson diversity based fuzzy clustering model (QWGSD-FC), is first proposed as a basis of our work. Based on QWGSD-FC and inspired by transfer learning, two types of cross-domain, soft-partition clustering frameworks and their corresponding algorithms, referred to as type-I/type-II knowledge-transfer-oriented c-means (TI-KT-CM and TII-KT-CM), are subsequently presented, respectively. The primary contributions of our work are four-fold: (1) The delicate QWGSD-FC model inherits the most merits of FCM, MEC and FC-QR. With the weight factors in the form of quadratic memberships, similar to FCM, it can more effectively calculate the total intra-cluster deviation than the linear form recruited in MEC and FC-QR. Meanwhile, via Gini-Simpson diversity index, like Shannon entropy in MEC, and equivalent to the quadratic regularization in FC-QR, QWGSD-FC is prone to achieving the unbiased probability assignments, (2) owing to the reference knowledge from the source domain, both TI-KT-CM and TII-KT-CM demonstrate high clustering effectiveness as well as strong parameter robustness in the target domain, (3) TI-KT-CM refers merely to the historical cluster centroids, whereas TII-KT-CM simultaneously uses the historical cluster centroids and their associated fuzzy memberships as the reference. This indicates that TII-KT-CM features more comprehensive knowledge learning capability than TI-KT-CM and TII-KT-CM consequently exhibits more perfect cross-domain clustering performance and (4) neither the historical cluster centroids nor the historical cluster centroid based fuzzy memberships involved in TI-KT-CM or TII-KT-CM can be inversely mapped into the raw data. This means that both TI-KT-CM and TII-KT-CM can work without disclosing the original data in the source domain, i.e. they are of good privacy protection for the source domain. In addition, the convergence analyses regarding both TI-KT-CM and TII-KT-CM are conducted in our research. The experimental studies thoroughly evaluated and demonstrated our contributions on both synthetic and real-life data scenarios. The novel QWGSD-FC model is designed for soft-partition clustering.Two cross-domain clustering methods: TI-/TII-KT-CM are subsequently proposed.TI-/TII-KT-CM are superior to or competitive with other existing methods.The convergences of both TI-KT-CM and TII-KT-CM are proved.

At present, in the field of person reidentification (re-id), the commonly used supervised learning algorithms require a large amount of labeled samples, which is not conducive to the model promotion. On the other hand, the accuracy of unsupervised learning algorithms is lower than supervised algorithms due to the lack of discriminant information. To address these issues, we make use of a small amount of labeled samples to add discriminant information in the basic dictionary learning. Moreover, the sparse coefficients of dictionary learning are decomposed into a projection problem of the original features, and the projection matrix is trained by labeled samples, which is transformed into a metric learning problem. It thus integrates the advantages of the two methods through combining dictionary learning and metric learning. After the data are trained, a projection matrix is used to project the unlabeled features into a feature subspace and the labels of the samples are reconstructed. The semisupervised learning problem is then transformed to a supervised learning problem with a graph regularization term. Experiments on different public pedestrian datasets, such as VIPeR, PRID, iLIDS, and CUHK01, show that the recognition accuracy of our method is better than some other existing person re-id methods.

A discriminative deep transfer metric learning method called DDTML is proposed for cross-scenario person re-identification (Re-ID). To develop the Re-ID model in a new scenario, a large number of pairwise cross-camera-view person images are deemed necessary. However, this work is very expensive due to both monetary cost and labeling time. In order to solve this problem, a DDTML for cross-scenario Re-ID is proposed using the transferring data in other scenarios to help build a Re-ID model in a new scenario. Specifically, to measure distribution difference across scenarios, a maximum mean discrepancy based on class distribution called MMDCD is proposed by embedding the discriminative information of data into the concept of the maximum mean discrepancy. Unlike most metric learning methods, which usually learn a linear distance to project data into the feature space, DDTML uses a deep neural network to develop the multilayers nonlinear transformations for learning the nonlinear distance metric, while DDTML transfers discriminative information from the source domain to the target domain. By bedding the MMDCD criteria, DDTML minimizes the distribution divergence between the source domain and the target domain. Experimental results on widely used Re-ID datasets show the effectiveness of the proposed classifiers.

Phishing emails are a real threat to internet communication and web economy. In real-world emails datasets, data are predominately composed of ham samples with only a small percentage of phishing ones. Standard Support Vector Machine (SVM) could produce suboptimal results in filtering phishing emails, and it often requires much time to perform the classification for large data sets. In this paper, an online version of imbalanced SVM (OISVM) is proposed. First an email is converted into 20 features which are well selected based on its content and link characters. Second, OISVM is developed to optimize the classification accuracy and reduce computation time, which is used a novel method to adjust the separation hyperplane of imbalanced date sets and an online algorithm to make the retaining process much fast. Compared to the existing methods, the experimental results show that OISVM can achieve significantly using a proposed expressive evaluation method. DOI : http://dx.doi.org/10.11591/telkomnika.v12i6.4562 Full Text: PDF

In medical datasets classification, support vector machine (SVM) is considered to be one of the most successful methods. However, most of the real-world medical datasets usually contain some outliers/noise and data often have class imbalance problems. In this paper, a fuzzy support machine (FSVM) for the class imbalance problem (called FSVM-CIP) is presented, which can be seen as a modified class of FSVM by extending manifold regularization and assigning two misclassification costs for two classes. The proposed FSVM-CIP can be used to handle the class imbalance problem in the presence of outliers/noise, and enhance the locality maximum margin. Five real-world medical datasets, breast, heart, hepatitis, BUPA liver, and pima diabetes, from the UCI medical database are employed to illustrate the method presented in this paper. Experimental results on these datasets show the outperformed or comparable effectiveness of FSVM-CIP.

As an emerging and challenging learning technique, group probability classifier learning is used to train a classifier from a group probability dataset, where the class labels of each sample are unknown, but the probabilities of each class in the given data groups of the whole dataset are available. The existing work is mainly based on the inverse calibration (IC) strategy to obtain the estimated labels for data in the group probability dataset and then use the classical classification algorithms to train the desired classifier. A critical challenging for the exiting IC based method is that an idea IC function for the label estimation is not easy to design, which makes the existing methods more sensitive to the adopted IC function. In order to overcome this shortcoming, a novel probability transductive classifier is proposed without the need of IC in the learning procedure, where the probability values are directly used as the output of the training data for the model training. Particularly, on the training data with the output as the continuous real values, the existing classical regression model can be easily used to model the group probability classification problem. For a future testing data, the model output of the obtained group probability classification model will present the probability of the testing data to the positive class. The T with the given threshold, the explicit class label of the test data is obtained for classification task. The experimental results on simulated datasets and real UCI datasets show that the proposed method is more effective compared with the existing methods.