Search

The digital trunking system has the characteristics that the calling mode is mainly group calling and the communication mode is mostly half duplex. For the problem that the traditional user classification method couldn't meet the classification requirements of digital cluster users, this paper proposed a digital cluster user classification method based on improved RFM model. Firstly, it introduced the average speech duration attribute to establish the recency vitality speak (RVS) model. Then, it used the analytic hierarchy process to determine the weight of each parameter in the model. Finally, it used the K-means + + clustering algorithm to classify digital cluster users. The simulation result shows that, by using the user classification method proposed in this paper, the accuracy of digital cluster user classification can reach more than 87. 9% . [ABSTRACT FROM AUTHOR]

Consensus algorithm based on credit has significant advantages in consensus efficiency and security, which has received great attention from academy and industry in recent years. This paper made a comparative analysis of the current consensus algorithms based on credit to provide reference for researchers to study in depth. Firstly, this paper systematically reviewed the state-of-the-art consensus algorithms based on credit, gave its development history, classification and core working principle, and summarized its credit calculation model and performance evaluation indicators. Secondly, it analyzed the creditbased consensus algorithms qualitatively and quantitatively, and pointed out their advantages and disadvantages. Finally, aiming at the problems of insufficient credit incentive and low integration with the consensus process in the state-of-the-art consensus algorithms, this paper proposed to optimize from the credit incentive based on game theory and designing efficient communication topology, and gave the future research direction. [ABSTRACT FROM AUTHOR]

Weakly supervised temporal action localization (WTAL) aims to locate the start and end boundaries of action instances and identify the corresponding actions. Although the existing methods have made great progress, there are still problems of incomplete localization and missing detection of shorter motions. To this end, this paper proposed a localization method of feature mining and region enhancement (FMRE) Firstly it calculated the similarity score between video segments through the base branch, and aggregated the context information with this score to obtain a more differentiated segment classification score, further realizing the complete positioning of the action. Then, it added a enhance branch to dynamically up-sample action proposals with a shorter duration in the initial localization along the temporal dimension, and then utilized the multi-head self-attention mechanism to explicitly model the temporal structure between action proposals, which facilitated action localization with temporal dependencies and prevented missing detection of short actions. Finally, it constructed pseudo-labels of mutual supervision between the two branches to gradually improve the quality of action proposals during the training process. The algorithm achieves mAP of 70.3% and 40.7% detection performances on the THUMOS14 and ActivityNet1.3 datasets respectively, which proves the effectiveness of the proposed algorithm. [ABSTRACT FROM AUTHOR]

Aiming at the problem of low accuracy and poor robustness of Visual Simultaneous Localization and Mapping (VSLAM） in dynamic environment, This paper proposed an adaptive Dynamic SLAM（AD-SLAM）. The scene dynamic classification front-end based on epipolar geometry constraint perceives the dynamic changes of the working environment in real time. The dynamic feature culling in low dynamic environment based on geometric constraints and moving probability and the dynamic target processing in high dynamic environment based on semantic segmentation eliminates moving objects. Experimental verification on the computer vision dataset TUM shows that the proposed method can improve the accuracy and robustness of localization in the mapping process of mobile robots while ensuring the real-time performance of the algorithm. [ABSTRACT FROM AUTHOR]

In reality, the data generated in many fields usually have multiple classes and are imbalanced. In multi-class imbalance classification, problems such as class overlap, noise and multiple minority classes reduce the capability of classifiers, and effective solution of multi-class imbalance problem has become an important research topic in the field of machine learning and data mining. Based on the recent literature on multi-class imbalance classification methods, this paper analyzed and summarized both data preprocessing and algorithm-level classification methods, and conducted a detailed analysis of all algorithms in terms of advantages, disadvantages and data sets. The data preprocessing methods introduced oversampling, under-sampling, hybrid sampling and feature selection methods to compare the performance of the algorithms using the same datasets. In addition, the algorithm-level classification methods described and analyzed base classifier optimization, ensemble learning and multi-class decomposition techniques. Finally, this paper summarizes the future development directions of the multi-class imbalanced data classification research field. [ABSTRACT FROM AUTHOR]

Most of the existing multi-label learning techniques consider the correlation learning problem but ignore the inconsistency feature of structural properties, which can change the structural property of the label space obtained by function mapping and it can directly affect the classification performance. In order to deal with this problem, this paper proposed a multi-label classification algorithm based on structural property preserving policy and correlation learning policies. Firstly, the algorithm constructed a mapping function to realize the mapping between the feature space and the label space. Secondly, it used structural property preserving strategy based on feature data to reduce the difference between feature data caused by linear transformation. Finally, this paper introduced the correlation learning strategy based on label data to further optimize the parameters of the proposed algorithm to improve the classification performance. The experimental results through a group of benchmark instances show that compared with the other classical multi-label classification algorithm, the proposed algorithm is much superior in classification performance, and the algorithm has great effectiveness. [ABSTRACT FROM AUTHOR]

Partial label learning is a weakly supervised learning framework. In partial label learning, each instance is associa-ted with a set of candidate labels, and its ground-truth label is unknown to us during the training process. In order to eliminate the ambiguous of candidate labels, this paper put forward a novel partial label learning by semantic difference and manifold learning (PL-SDML) method, which combined the semantic difference maximization criterion of instances and manifold lear-ning for partial label learning. The PL-SDML method was a two-stage method that used semantic difference maximization criterion of instances and manifold learning to generate the labeling confidence for training instances in the training phase. Then, PL-SDML made predicts for unseen instances via a nearest neighbor voting-based approach in the predict phase. On the UCI datasets, PL-SDML is superior to other comparison algorithms in 70% cases. On the four real-world datasets, the classification performance of PL-SDML improves by 0.3%~13.8% compared with other baselines. [ABSTRACT FROM AUTHOR]

In a class-imbalanced dataset, since the samples close to the class boundary are more likely to be misclassified, it is of great significance to accurately identify boundary samples for classification. Existing methods usually use K-nearest neighbors to identify boundary samples, but the accuracy needs to be improved. To address the above problem, this paper proposed a synthetic oversampling method for boundary minority samples based on Tomek links. This method first found inter-class samples with the nearest distance to form Tomek links. Then, it identifies those minority samples located at the inter-class boundary according to Tomek links. Next, it used the linear interpolation mechanism in synthetic minority oversampling technology (SMOTE) to perform oversampling between the boundary samples and their minority neighbors, thereby achieving the balance of the datasets. The comparison experiment with eight sampling algorithms shows that the proposed method can obtain higher G-mean and F, values on most of the datasets. [ABSTRACT FROM AUTHOR]

The deep decision tree migration learning boosting method ( DtrBoost) can effectively realize the migration learning from a single source domain to a target domain under supervision, but can not realize the unsupervised migration scenario under multiple source domains. To solve this problem, this paper proposed an unsupervised transfer learning boosting method for optimizing the weight under multi-source domain distribution. The main idea was to calculate the corresponding KL value according to the distribution of different source domains and target domains, selected an appropriate number of samples from different source domains to train the classifier and pseudo label the samples from the target domain. Finally, the algorithm assigned different learning weights according to the KL distance of each different source domain, and the labeled source domain samples integrated to the pseudo labeled target domain to obtain the final result. Comparative experiments show that the proposed algorithm achieves better classification accuracy and adaptive effect on different data sets. The average classification error rate decreases by 2. 4% and more than 6% on the best marketing data set. [ABSTRACT FROM AUTHOR]

Aiming at the problem of fine-grained visual categorization, this paper proposed an effective algorithm to achieve end-to-end fine-grained visual categorization. The ECA module in ECA-Net was a channel attention mechanism with significant performance advantages. It model-fused with the classic network ResNet-50 to form the ResEca. Then, it used the object-level image positioning module and the part-level image generation module to generate object-level and part-level images. Those images combined with original images could be as the input of the new constructed network Tb-ResEca-Net. This paper trained the 95. 1% and 95. 3% respectively on the test set of the corresponding dataset. The experimental results show that this method has higher classification accuracy and stronger robustness compares with other traditional fine-grained classification methods, which is an effective fine-grained image classification method. [ABSTRACT FROM AUTHOR]

Using feature selection technology in data mining to define the key features from high-dimensional and complex omics data has been the focus of research. This paper studied the complex correlation between features of the omics data, and proposed a feature module selection algorithm based on the synergetic network. The interaction gain was used to construct the weighted synergetic network. By measuring the closeness of candidate nodes to the current feature module and combining with the node’s own classification performance, important feature module can be obtained. The algorithm was validated on ten public data sets and showed advantages over the comparison algorithms in classification accuracy, sensitivity and specificity, which indicates the feature modules selected by the algorithm have better performance. [ABSTRACT FROM AUTHOR]

The current ADMET classification methods have shortcomings in classifying the ADMET of compounds data with multiple characteristics and feature correlation. Moreover, the classification results of ADMET are not explainable. To solve these problems, this paper proposed a classification model based on capsule network（CapsMC). CapsMC model first proposed the Feature-to-Image algorithm. It used this algorithm to consider the correlation and dependence between features into the classification basis, and realized the multi-level extraction of features. Second, it explored the advanced application of capsule network, and designed a cognitive reasoning mechanism. It used this mechanism to carry out cognitive reasoning on features, and realized explainable classification of ADMET. Experimental results on five ADMET datasets show that CapsMC can achieve the explainable classification of ADMET well. [ABSTRACT FROM AUTHOR]

Representation-based classification ( RC) usually uses all the training samples of all pattern classes to represent a test sample . However, whether it is necessary to use all pattern classes to represent the test sample remains to be investigated . So, this paper proposed two-stage representation-based classification framework. First, it used a RC algorithm to calculate the representation coefficients of the test sample w. r. t the training samples of all classes, and found the first k ( k≥1) classes with the minimum representation error. Then with the training samples of these k classes, it applied a RC algorithm again to represent the test sample, based on which, the label of the test sample was finally determined by finding the minimal representation error class among those k classes. In addition, this paper developed a nonnegative weighted collaborative RC ( NWCRC) algorithm. The first and second RC algorithm of the proposed RC framework could be the same or different. Setting the first and second RC algorithms to be the same, with five different RC, experiments were carried out on five databases, experimental results of which show that the proposed two -stage RC framework can significantly improve the classification accuracy of the original RC algorithms in most cases. [ABSTRACT FROM AUTHOR]

The effective classification of cloud manufacturing resources is a prerequisite for resource service packaging and subsequent manufacturing resource service retrieval. And it aims for the classification of high-dimensional features and multicategory processing equipment resources. At first this paper extracted resource attribute characteristic values and performed vectorization processing by XML documents which described the characte ristics of cloud manufacturing resource attributes . Then, based on the SVM algorithm, it established an automatic classification and prediction algorithm model for manufacturing resources. In order to improve the classification efficiency of the SVM classification algorithm, this paper introduced the SMO algorithm, and it used the grid search to find the optimal parameters of the SVM algorithm for optimizing the effect of the SVM algorithm. Finally, based on the original equipment information of a certain equipment resource set, the paper conducted a classification experiment on different processing equipment to verify the validity and feasibility of the classification model. [ABSTRACT FROM AUTHOR]

The aim of target speaker ' s speech separation is to extract one's speech from a mixture speech consisted of multiple speakers, which is guided by an eigenvector. There are two ways to get the eigenvector, one is to use a one-hot vector, another is to adaptively generate an embedding containing the target speaker ' s characteristic form a classification neural network. The advantage of using one-hot vector is that it can achieve perfect performance during the training process, while it cannot handle the unseen speakers beyond the training set. The advantage of using embedded vectors is that it loses part of the training effect,but it has a good generalization effect on unseen speakers. In order to solve the shortcomings of singlehandedly using onehot vector or embedding vector in specific speaker speech separation algorithm, this paper proposed a hybrid training method. It used the one-hot vector and embedding vector alternately as the identity feature vector of target speakers. By mapping one-hot and embedding into public space, the proposed method could achieve a good generation effect while ensuring the training effect. The experimental results show that the proposed method achieves more than 10 dB SDR improvement on unseen speakers' speech separation. [ABSTRACT FROM AUTHOR]

Most existing positive-unlabeled (PU) graph learning methods exact only node representations to infer node labels in dependently. This paper proposed a collective inference based method that exploited the correlations among nodes to assist in classification of unlabeled nodes. Firstly, it used the personalized PageRank algorithm to approximate correlation degrees between each node and observed positive nodes as a whole (positive correlation degree, PCD for short). Then, it built a local classifier to predict labels of unknown nodes by combining PCD with node representations that were captured via a graph neural network (CNN), and constructed a relational classifier to collectively up date labels of unknown nodes by combining PCD with local label dependency of nodes that were exacted via another CNN. Furtherly, it exploited the Markov CNN (CMNN ) frame-work to train these two classifiers, alternately and iteratively, to form a multi-hop collective inference procedure. Besides, it proposed a mixed non-negative unbiased risk estimator for the two classifiers to estimate empirical loss with only positive and unlabeled nodes. Finally, either of them could predict labels of unknown nodes. Experimental results on real-life datasets show that the proposed method remarkably out performs the state-of-the-art approaches in identifying both single-class target concept and multiple-classes-merged target concept, and performs quite robust against to error of the prior of positive nodes. [ABSTRACT FROM AUTHOR]

The development of society has brought countless data, with the unbalancedness becoming a significant feature of many data sets. So it has come to be a research hotspot for machine learning on how to make those unbalanced data sets obtain better effects of classification. Based on this, this paper conducted a comprehensive research on the current unbalanced data set classification method, and made an overall interpretation and conclusion from such three aspects as the unbalanced data sampling method, the method of machine learning-based improved algorithm and the combination method. It also analyzed and took int o account many factors, including the problems solved by each method, algorithm mentality, application scenarios, as well as the advantages and disadvantages of each, and delivered a summary on potential problems of the classification methods and a prospect on the future research directions [ABSTRACT FROM AUTHOR]

For binary classification problems, the classifier based on the discriminant model usually searches for an optimal decision boundary, which is susceptible to data fluctuations. This paper proposed a Q-learning algorithm based on the generative model for binary classification ( BGQ-learning), which coded state and action separately and obtained corresponding decision functions, increasing the flexibility of decision space. And then it combined least squares temporal-difference (TD) algorithm and semi-gradient descent for parameter optimization, accelerating parameter convergence speed. This paper designed experiments to compare the performance of the proposed algorithm with three classical classifiers and a novel classifier. The test results on 7 data sets of the UCI database show that the proposed algorithm has excellent stability and classification accuracy. [ABSTRACT FROM AUTHOR]

The paper discussed classification limitation and computational complexity of categorization algorithm based on sparse representation, and proposed a novel weighted locally linear KNN algorithm for text categorization, which used weighted function to make the representation coefficients sparse, and introduced nonnegative constraints to improve the classification performances. Moreover, this paper gave the parameters experiments to select the optimization value, and gave the theoretical proof of the algorithm. The experiments show the superiority of the algorithm based on weighted locally linear KNN, which has an average improvement on 2. 49% and 0. 85% in precision and recall, compared with the traditional mcxlel. It means the text categorization algorithm based on weighted locally linear KNN has advantages of high classification accuracy and strong convergence, which is suitable for high-dimensional data classification. [ABSTRACT FROM AUTHOR]

Object tracking is an important topic in computer visual directions, where scale changes, deformation and rotation are difficult to resolve in the field. For the challenging problems faced in the above track, based on existing Siamese network algorithms, this paper proposed multi-layer feature fusion and parallel self-attention Siamese networks (MPSiamRPN) for visual tracking. Firstly, MPSiamRPN used the modified ResNet50 to extract features from the template image and the search image. In order to deal with the loss of some features caused by the deep network, it proposed a multi-layer feature fusion module to fuse the features of the last three layers of ResNet. Secondly, it introduced the parallel self-attention module, which was composed of channel self-attention and spatial self-attention. The channel self-attention could selectively emphasize the beneficial channel features for tracking, and the spatial self-attention could learn the rich spatial information of the target. Finally, it proposed the region proposal network ( RPN) to perform classification and regression operations to determine the location and shape of the target. Experiments show that the MPSiamRPN can achieve competitive results on OTBlOO and VOT2018 test datasets. [ABSTRACT FROM AUTHOR]

DTW is one of the commonly used algorithms in time series similarity measurement. However, DTW has the shortcoming of pathological alignment and ignores the influence of time attribute. LDTW and TDTW have been proposed to handle two shortcomings of DTW separately, however they cannot be solved simultaneously by LDTW or TDTW independently. This paper proposed TLDTW algorithm. Firstly, it constructed time weight matrix by measuring the distance between points in two series. Secondly, it fused the corresponding time weights from time weight matrix into the recursive filling procedure for cumulative cost matrix of LDTW, thus it considered the time attribute and the problem of pathological alignment could still be suppressed. It conducted 1-NN classification experiment based on UCR dataset, and experimental results show that the classification accuracy based on TLDTW is better than other compared algorithms, and the reliability of TLDTW is verified by further comparison. [ABSTRACT FROM AUTHOR]

The traditional naive Bayes classification algorithm does not divide the importance degree according to the different feature items, which makes the classification result inaccurate. In order to solve this problem, this paper introduced Jensen-Shannon(JS) divergence and used JS divergence to express the amount of information provided by the feature terms. Aiming at the deficiency of JS divergence, the paper considered from the three aspects of word frequency, text frequency and inverse category frequency corrected by coefficient of variation, and then adjusted and corrected the JS divergence. Finally, it calculated the weight of each feature and introduced the weights into the naive Bayes formula. Compared with other algorithms, it proves that this method improves the naive Bayes classification algorithm effectively. Therefore, compared with other classification algorithms, the performance of naive Bayes classification algorithm based on JS divergence feature weighting is greatly improved. [ABSTRACT FROM AUTHOR]

Aiming at the problem that traditional vulnerability detection classification requires the definition of artificial features, similarity matching algorithms cannot detect non-clonal vulnerabilities and there are large feature dimensions and only for function call in existing deep learning vulnerability detection methods, this paper proposed a deep learning method based on sliding window and hash function to perform static vulnerability detection and classification on source code. Firstly, it extracted the method body of the source code to form a positive and negative sample set constructed an abstract syntax tree for each sample, replaced the programmer-defined variable names and method names according to the node type in the syntax tree and serialized abstract syntax tree by preorder traversal. Then, it performed word segmentation on the node information in the abstract syntax tree node and assigned an independent node number for each word. Then, it further split the tree nodes to form a word sequence, and trained the vulnerability detection classification model based the sliding window and hash function. Finally, it selected two types of vulnerability data sets, i.e. CWE-190 and CWE-191,for experiments in the SARD data set. The accuracy and recall rate of the vulnerability detection classification model reach 97. 4% and 94. 2% for CWE-190 and 97. 6% and 95. 1 % for CWE-191 respectively. The results show that the model can effectively detect the types of security vulnerabilities in the code and it is superior to some existing methods. [ABSTRACT FROM AUTHOR]

The software of digital PCR system lacked the function for classifying droplets of multiplex experiments. This paper proposed a grid regional DBSCAN clustering algorithm. Firstly, it mapped data to the gridded space and established gridded indexes ;then it calculated weights of grids and performing deep search expansion;finally,it mapped the clustering results in gridded space back to the data space . Tests on artificial datasets show that the proposed algorithm is able to identify the sample points in edge area and of high efficiency. On this basis, it proposed a classification method of digital PCR droplets based on grid regional DBSCAN clustering. It used both the proposed method and existing methods to classify digital PCR droplets. The results show that this method is a convenient and accurate method for classification of digital PCR droplets. Therefore, the proposed method is suitable for classification of digital PCR droplets. [ABSTRACT FROM AUTHOR]

In order to improve the classification performance and operation efficiency of the hash algorithm, this paper pro· posed an image hash algorithm based on gradient change feature and energy features. Firstly, the algorithm preprocessed the input image to obtain a secondary image, and used the Sobel operator to calculate the x-axis and y-axis gradient values of the red channel, green channel, and blue channel images of the secondary image . It added the gradient values of the images to obtain the final gradient image. Then it used the multi-directional change information of the gradient image amplitude as the gra-client feature, and used the energy values of all image sub-blocks as the energy features. Finally, it combined and scrambled the gradient feature and energy features of the image to generate the final hash sequence. The experimental results show that : the discrimination and robustness of the proposed algorithm can achieve a good trade off; compared with the latest and excellent hash algorithms, this algorithm has best ROC curve and the shortest average running time 0. 0242 s, and in the comparison experiment of copy detection, the R-P curve of the proposed algorithm is the best. [ABSTRACT FROM AUTHOR]

Deep neural networks are vulnerable to the attack of adversarial examples.To solve this problem,some works trained networks by adding Gaussian noise to the image,thereby improving the ability of the network to defend adversarial examples.But the method did not consider the sensitivity of the network to different areas in the image when adding noise.To solve this problem,this paper proposed an adversarial training algorithm based on gradient guidance noise addition.When training the network,it added adaptive noise to different areas based on the sensitivity,added large noise to the more sensitive areas,suppressed the sensitivity of the network to image changes,added less noise to the less sensitive areas and improved the network classification accuracy.Compared with the existing algorithms on the Cifar-10 dataset,the experimental results show that the proposed method effectively improves the accuracy of neural networks when classifying adversarial examples. [ABSTRACT FROM AUTHOR]

To filter out the false positive patterns returned from the discriminative sequential patterns mining methods,this paper proposed a standard permutation method called SP-DSP.This method firstly mined frequent sequential patterns by the GSP algorithm,then eliminated the patterns whose growth rate were less than the threshold.Finally,it used the standard permutation method to compute the p-values of tested patterns.As a result,the number of false positive patterns could be controlled under the multiple hypothesis testing measures.The experiments show that the SP-DSP algorithm can alleviate a lot of false positive patterns and retain as many true patterns as possible,which improves the accuracy of the downstream classification tasks. [ABSTRACT FROM AUTHOR]

Aiming at the problems of numerous complaint factors and relatively poor fault samples of Internet protocol TV users, this paper proposed the LFOA-HSRVM algorithm as a prediction method for IPTV user's complaint combined with the outstanding sparsity of relevance vector machine . The method regarded fault prediction of IPTV user as a binary-classification problem with imbalanced data sets. The algorithm was based on the nuclear parameter optimization of RVM and hybrid sampling technology to overcome the problem that the decision boundary of the traditional RVM algorithm was biased to the minority class wh en dealing with imbalanced data sets. Compared with other algorithms, the experimental results show that the algorithm ' s performance of few classification and overall classification are improved greatly, and the effect of prediction is better. [ABSTRACT FROM AUTHOR]

Due to the high-speed requirement of high-end network devices, hardware using TCAMs has been the dominant implementation of packet classification in industry. Despite its capability for line-speed queries, TCAM is not only power hungry but also capacity inefficient. By making use of a pre-classifier to activate TCAM blocks selectively, many research efforts significantly reduce the power consumption of TCAM. However, these bottom-up based pre-classifiers achieve power savings at the expense of poor utilization of TCAM capacity, and the potential of power reduction is not fully exploited in many cases. This paper proposed a power-saving pre-classifier for TCAM based packet classification, which constructed based on decision trees. By grouping rules with respect to their small fields, rules could be recursively mapped into decision trees without the trouble of rule replications, so that a top-down traversal algorithm could be well applied for obtaining index items. Experimental results show that for rule sets up to one hundred thousand entries, the proposed design achieves 98. 2% power reduction with a TCAM storage overhead of 1. 3% on average. [ABSTRACT FROM AUTHOR]

Aiming at the problem of the discriminator' s poor classification accuracy on a small number of labeled samples and insufficient robustness to the local perturbation of manifolds in the generative adversarial network, this paper proposed a generative adversarial network based on variable loss and manifold regularization. The algorithm used a variable loss instead of the original discriminator to solve the adverse effect of the poorly trained classifier on the semi-supervised classification task. In addition, on the basis of variable loss of discriminator, it added manifold regular terms to improve the robustness of discriminator to local disturbance by punishing the variation of classification decision of discriminator on manifold. Using the quality of the generated samples and the semi-supervised classification accuracy as the evaluation criteria of the algorithm, it performed numerical experiments on the dataset SVHN and CIFAR-10. Comparing with other semi-supervised algorithms, the results show that the proposed algorithm can obtain higher quality generated samples and higher precision classification results with a small amount of labeled data. [ABSTRACT FROM AUTHOR]

The traditional measurement method of similarity ignores the project of multi-attribute category differences. To avoid this problem, this paper proposed a recommendation algorithm for improving project multi-attribute classification. Firstly, it used the project-user membership matrix to explore the affiliation and created a similar neighbor FP-Tree to extract the nearest neighbor set. Then it analyzed the common item similarity between users and the difference of the project multi-attribute classification, and used the weight factor to combine the common project with multi-attribute classification. It constructed the CNB model to measure the similarity degree of neighbors. Finally, it sorted the similar users in descending order to obtain more accurate similar users and complete the recommendation work. In virtue of the medical dataset, it verified the effectiveness of the proposed algorithm. The experimental results show that the recommendation accuracy of time complexity and mean of average accuracy have been improved. [ABSTRACT FROM AUTHOR]

At present, most intrusion detection algorithm cannot achieve a good balance between intrusion detection rate and false positive rate, in order to effectively avoid such problems, this paper proposed an intrusion detection method based on ADEN. The method first initialized the parameters of the encoder part in the ADEN model by training the deep belief network, and initialized the parameters of the decoder part by using the normal distribution. Then it tuned the parameters of the asymmetric deep belief network model by calculating the reconstruction error, so that the model can obtain the optimal low-dimensional representation of the original data. Finally, it used the data obtained by the encoder as input data of the classifier and detected. The ADEN model can extract features that are more conducive to classification and save more test time in the model initialization phase. The experimental results show that the method can achieve better detection performance and achieve better detection accuracy for small categories of samples. [ABSTRACT FROM AUTHOR]

In order to solve the problem of classification of large-scale intrusion data, this paper proposed lightGBM binary tree algorithm based on stacked sparse autoencoder. Firstly, it divided the category labels into five categories and constructed into binary tree structures. Then solved the imbalance of data distribution by the upper sampling method, the above processing could separate the large-scale data, so that they could be trained separately. Next, it used the sparse autoencoder network to reduce the feature dimension. Using this method could ensure that time of dimension reduction could save on the basis of extracting deeper features from the original data. Finally, it used the lightGBM ensemble algorithm to classify. And compared to other models, using the lightGBM model could save training time while ensuring classification performance. It used the NSL-KDD dataset to measure the accuracy, precision, recall. And comprehensive evaluation index F1 of the proposed method, which reached an average of 87. 42%, 98. 20% and 91. 31% in five classification, respectively. It is superior to the comparison algorithm and obviously saves the calculation time . [ABSTRACT FROM AUTHOR]

Aiming at the problems of low recall rate and poor running efficiency caused by traditional abnormal accounts detecting algorithms in non-balanced social network datasets, the paper extracted user content, behavior, attributes, and relationship features from social network data sets, selected features using gradient-enhanced ensemble classifier XGBoost algorithm, established classification model,constructed unbalanced data sets and realized the identification of three types of spam accounts. Experimental results show that it improves the recall rate and the F1 value in identification of three types of abnormal users effectively by XGBoost algorithm in binary classification and multiple classification tasks both in the balanced and unbalanced dataset in comparison with the traditional classification methods such as random forest. And with few features selected by XGBoost, the classification algorithm can get the same effect as with all features of samples, which proves the effectiveness of this method. [ABSTRACT FROM AUTHOR]

Aiming at the challenges faced by the existing big data-oriented computing framework in the study of extensible machine learning, this paper proposed a distributed naive Bayesian text classification method based on MapReduce and Apache Spark framework. This method explored the Bayesian network classifier by studying the adaptability of MapReduce and Apache Spark frameworks, and studied the existing computing framework for big data. First, it divided the training sample data set into m classes based on the naive Bayes text classification model. In the training phase, it used the output of the previous MapReduce as the input of the next MapReduce, and used four MapReduce jobs to derive the model. This design process made full use of the parallel advantages of MapReduce. Finally, when the classifier was tested, it obtained the value of the class label which the maximum value belonged. Experiments in the Newgroup's dataset show the proposed method achieves more than 99% of the results on all five types of news data sets, and is all higher than the comparison algorithms, which prove the accuracy of the method. [ABSTRACT FROM AUTHOR]

Many malware families such as botnet utilize domain generation algorithms (DGAs) to evade detection at present. The mainstream detection algorithms based on artificial rules and machine learning have some limitations due to the fact that DGAs generate domain characters timely and rapidly. The former is somewhat blind to new DGA variants. The latter suffers from the lack of evolving training data. In order to solve these problems, this paper defined domain encoder and decoder on account of the method of ASCII encoding and combined them with the concept of generative adversarial network (GAN) to construct domain character generator. Then it used the generator to predict and generate DGA variants. Experiment results show that the DGA variants generated by this method can act as real DGA samples when these variants are utilized to train and estimate classifiers. This verifies the validity of the generated data and they can be effectively utilized to train and estimate DGA domain detector. [ABSTRACT FROM AUTHOR]

Aiming at the reasonable placement of different business contents in content centric networking (CCN), this paper proposed a hybrid caching scheme based on service classification and node partition (SCNP). For streaming media-on-de-mand services, this scheme used push-pull cache strategy based on popularity to achieve its sequential storage in the edge network. For non-streaming contents, this scheme used hash-based explicit cache strategy to implement its placement of single copy in the core network. Simulation results show that compared with the classic cache algorithms, this scheme can improve the cache hit ratio and the hop reduction ratio, and reduce the average request delay. [ABSTRACT FROM AUTHOR]

Aiming at the low precision problem while the cost was introduced into attribute reduction of decision-theoretic rough set, this paper studied the balance between the total cost and the precision in classification. It used the total cost of the classification and the approximate classification quality as the constrained criteria in the attribute reduction procedure, combined with simulated annealing method, and proposed a DTRS attribute reduction algorithm constrained by cost-sensitive and classification quality (hereinafter referred as ARACOQ). It carried out the simulation experiments by using UC1 data set. The results verify the effectiveness of the ARACOQ algorithm. which can find an attribute reduction set with the highest classification precision within the affordable cost range. [ABSTRACT FROM AUTHOR]

Aiming at optimizing and improving a hybrid collaborative filtering based on Spark platform for its sparsity, scalability and personalized recommendation by using the method of algorithm integration, this paper took the model of Stacking integration to integrate multiple weak recommender units in a linearly weighted into a comprehensive recommender. Firstly, this algorithm optimized the collaborative filtering based on the nearest neighbor by presorting and adjusting the similarity calculation strategy with popularity and praise degree, and improved the rationality and complexity of similarity calculation. It solved the problem of score sparsity to some extent. At the same time, this algorithm integrated closely distributed computing platform, which could make full use of the advantages of distributed platform to design and implement an increment iterative model of recommendation algorithm by using the Spark streaming and distributed storage structure. It solved the problem that collaborative filtering algorithm was hard to expand and made poor real-time performance. The experimental data used UCI public data set named MovieLens and NetFlix films’ score. The experimental results show that the improved algorithm has a good performance and makes great progress in personalized recommendation, accuracy and scalability compared with the previous algorithms. It provides a feasible algorithm integration scheme for the application of the recommended system. [ABSTRACT FROM AUTHOR]

Multi-label learning has been widely used in many application scenarios right now. In this kind of learning problem, each instance is simultaneously assigned with more than one class label. Since different class labels might had their own unique characteristics(such as label-specific feature) which would be more useful for label classification, so some multi-label learning approaches based on label-specific features had already been proposed. Therefore, aiming at the problem that redundant feature space caused by label-specific feature construction, this paper proposed a multi-label label-specific feature extraction algorithm named LIFT_RSM, which could improve the performance of classification by comprehensively using random subspace method and the thought of pair-wise constraint dimensionality reduction to extract effective feature information in labelspecific feature space. The experimental results on several datasets show that the proposed algorithm can achieve better classification results compared with several classical multi-label algorithms. [ABSTRACT FROM AUTHOR]

The application of neighborhood rough set depends on the value of neighborhood size δ. When using attribute reduction algorithms based on neighborhood rough set, an existing method for determining δ is usually point-type, that is, to specify a value only by human experience. The method does not combine with the actual situation when it is used to determineδ, so the practicability of the algorithms can be further discussed. For this reason, this paper proposed an adaptable method for determining δ, which the biggest characteristic was not determining δ but the interval of δ. It forwardly selected the most appropriate δ in the interval by using a fitness function that was combined with the characteristics of data sets and classifiers. The experimental results show that, compared with the point-type method for determining δ, this method can find reduction sets which number of attributes is less, and classification accuracy is higher. It proves that this method can further improve the practicability of attribute reduction algorithms based on neighborhood rough set. [ABSTRACT FROM AUTHOR]

In order to improve the classification accuracy of the minority samples, this paper proposed a novel resampling algorithm based on the improved density peaks clustering method, named IDP-SMOTE. First, it improved density peaks clustering algorithm by utilizing Box-Cox transformation and σ-rule for finding the clustering centers and outliers automatically. Second, it combined the improved density peaks clustering algorithm with SMOTE method. With removing the noisy data, the synthetic samples could be generated in the sub-class regions on the basis of the values of local density and nearest distance of the minority samples. The presented algorithm avoids the boundary ambiguity caused by over-sampling, improves the imbalance problem with-in class and reduces the learning difficulty of the boundary data. Meanwhile, it realizes automatic clustering and resampling, and avoids the interference of subjective factors. Through the contrast experiment, the proposed algorithm is effective and adaptive. [ABSTRACT FROM AUTHOR]

Relation recommendation based on common neighbors' property is a hot research branch of link prediction in social network analysis. This paper proposesd a new measure of relation recommendation by introducing a hidden naive Bayesians (HNB) classification model, which model the task by analyzing the dependency among properties and incorporates this idea to measure the influence and contribution among common neighbors. Then built a ranking model to learn the highest similarity associated with each candidate pair by maximizing the likelihood of relationship building and extended the model to CN, AA and RA similarity-based recommendation algorithms. Experimental evaluation by AUC on real social networks proved that the proposed model can achieve a better result than some baseline and LNB. Finally, it also discovered that attributes with different network topologies recommended precision linear effects. [ABSTRACT FROM AUTHOR]

Ensemble learning by integrating multiple base classifiers that trained different set can effectively improve the classification accuracy. In the base classifier training process, imbalanced data set can be processed by either cost-sensitive or data sampling technology. Due to the advantages of ensemble learning in imbalanced data classification, ensemble algorithms for imbalanced data classification have been widely research. This paper surveyed the state of the art of imbalanced data ensemble classification algorithms, including the mechanisms and features of major existing learning algorithms, their advantages and disadvantages, highlighted the open research issues and future research directions. [ABSTRACT FROM AUTHOR]