Search

Class imbalance poses difficulties in training a classifier that perform well on minority classes, especially when there is a high imbalance ratio and significant class overlap. Existing data-level methods often suffer from problems like information loss and overfitting. To address these problems, we introduce a novel majority affiliation based under-sampling method (MAUS). The MAUS method employs a support vector data description model to capture the distribution of the minority class, thereby forming a hyper-sphere to establish a majority affiliation for each sample. The high-dimensional hyper-sphere constructed through all minority class samples avoids the problem of overfitting. Leveraging the majority affiliation in conjunction with the k-nearest neighbor algorithm, MAUS is capable of identifying region of class overlap and subsequently removing majority samples within these regions that negatively impact classification performance. This selective removal process minimizes excessive information loss at classification boundaries while alleviating the issue of class overlap. Furthermore, by removing those majority samples that are situated far from the classification boundary, MAUS reduces the imbalance ratio to the expected value, resulting in the attainment of a balanced dataset. To validate the effectiveness of our method, we conducted extensive experiments comparing it with state-of-the-art methods on 30 publicly available datasets. The results indicate that our approach outperforms existing methods on most of datasets and classifiers. • Based on hyper-sphere built by all minority samples, majority affiliation quantifies the probability of a sample belonging to the majority class. • Each k-nearest neighbor majority sample within the overlapping region that adversely impacts the classification performance were removed. • Constrained by the expected imbalance ratio, majority sample with the largest weight is removed in turn, ultimately yielding a balanced dataset. • Two sets of experiments were conducted on 30 real-world datasets to evaluate its performance against 13 state-of-the-art resampling methods. [ABSTRACT FROM AUTHOR]

In recent years, the coexistence of different states in the neural system has attracted widespread interest. Researchers have found a coexisting state of spiking and resting in homogeneous networks, which is known as the chimera-like state. The real cortical network is a much more complex and heterogeneous network. Therefore, the excitatory-inhibitory cortical neuronal network is constructed based on Hodgkin-Huxley neuronal model in this paper, and the chimera-like state is further investigated in the heterogeneous network. It is found that the chimera-like state is related to the balance between excitatory and inhibitory synaptic currents. The excitatory coupling current can counteract the initial condition effect and promote synchronized firing of neurons in the network. The inhibitory coupling current desynchronizes the network and thus induces synaptic noise, resulting in an inverse bell-shaped dependence of the change in the number of spiking neurons. We analyzed the underlying mechanisms of synaptic noise in the phase plane diagram and found it has asymmetry for the neuronal state transition. In addition, neurons with low degrees have a higher probability of undergoing state transitions. Finally, we verified that the chimera-like state is robust to network topology and initial conditions. The results provide a new insight into neuronal interactions in heterogeneous networks and might help to reveal the mechanisms of coexistence of different states in the cortical network. • A excitatory-inhibitory cortical neuronal network model is proposed to study the chimera-like state. • The existence of the chimera-like state is related to the balance between excitatory and inhibitory synaptic currents. • The inhibitory coupling strength can lead to network desynchronization and thus induces synaptic noise. • Neurons in the network have a higher probability of state transitions at low in-degree. [ABSTRACT FROM AUTHOR]

Community structure is an important characteristic of complex networks. It determines where important functions of a network are located. Recently, discovering community structure in complex networks has become a hot topic of research. However, the continuous increase in network size has made network structure more complex, and community detection has become extremely difficult in real applications. In particular, the detection results are usually not accurate enough when classical clustering methods are applied to high-dimensional data matrices. In this paper, inspired by the relationship between vertices, we design a novel and effective network adjacency matrix transformation method to describe vertices' similarity in the network topology. On this basis, we propose a framework to extract nonlinear features: community detection with deep transitive autoencoder (CDDTA). This framework can obtain powerful nonlinear features of a real network to make community detection algorithms perform excellently in practice. We further incorporate unsupervised transfer learning into the CDDTA (Transfer-CDDTA) by minimizing the Kullback–Leibler divergence of embedded instances, to discover powerful low-dimensional representations. Finally, we propose a new training strategy and optimization method for our algorithm. Extensive experimental results indicate that our new framework can ensure good performance on both real-world networks and artificial benchmark networks, which outperforms most of the state-of-the-art methods for community detection in social networks. [ABSTRACT FROM AUTHOR]

When analyzing graph-structured data, graph neural networks (GNNs) use well-designed propagation mechanism to extract node features. However, these methods only incorporate data from nodes' fixed neighborhoods or whole graph, not accounting for the influence of important nodes outside of the neighborhoods. We offer transferable graph neural networks with deep alignment attention (TGDAA), which is a novel neural network framework that designs aligned attention to assign different weights to related important nodes. In such a scenario, we concurrently remove a number of graph neural network obstacles and make TGDAA accessible for transfer learning issues. Extensive experiments in four aspects: traditional metrics, attention-related methods, varying label rates and ablation study show that the suggested TGDAA performs better than ten state-of-the-art techniques on ten benchmark network datasets. [ABSTRACT FROM AUTHOR]

With the rapid development of online advertising, the click-through rate (CTR) prediction plays an important role in improving the benefits of advertising and user experience. CTR is the most commonly used evaluation indicator of the effects of online advertising. At present, the keys including feature extraction and user click behavior modeling have been taken into consideration by many researchers to design methods for CTR prediction. However, the characteristics of high-dimensional data sparseness and imbalance in advertising data are not fully considered, which results in the insufficient utilization of advertising information. To alleviate the problems of data sparsity and imbalance, this paper proposes a robust integrated locally kernel embedding (RILKE) model to solve data sparseness and incorporate unsupervised transfer learning into RILKE to form an improved model, named robust transfer integrated locally kernel embedding (RTILKE). Through theoretical analysis and empirical experiments, RTILKE can efficiently solve the data-imbalance problem of CTR prediction in online advertising, which effectively improves the prediction performance of advertising responses. [ABSTRACT FROM AUTHOR]

Purpose Thermal stress of the rotor in a squirrel cage induction motor is generated due to the temperature rise, and the structure of the rotor will be destroyed if the stress acted on the rotor exceeds its limits, so the thermal stress is also one of the main causes led to broken bar fault. The purpose of this paper is to report the thermal stress coupled analysis for the induction motor with healthy and faulty rotor, and to find the variation tendency of the temperature and thermal stress due to broken bars, and the part most likely to break in the rotor as a result of the thermal stress load are identified.Design/methodology/approach The steady temperature and thermal stress of the rotor in the case of the healthy and faulty conditions are calculated by finite element method, and the 3D model of the motor used in the experiments is established and the experimental results are presented for both healthy and faulty machines.Findings The influence of the broken bars fault on the motor thermal profile and thermal stress can be found, and it explains why the breaking point always appears in the joint of the bars and end rings.Originality/value The paper presents the 3D thermal stress coupled model and performance characteristics of induction motor with broken bars. The reasonable constraint is established according to the contact of components each other, and more reasonable fracture location is selected. The results obtained by the simulation model are in a good agreement with practical situation, because the effect of skewed rotor were taken into consideration in the process of simulation. [ABSTRACT FROM AUTHOR]

Abstract: Prototype Reasoning using granular objects is an important technology for knowledge discovery. Fuzzy clustering can be used to generate prototypes with different granularities. In order to find optimal granular prototypes through fuzzy clustering, for given data, two conditions are necessary: a good cluster validity function, which can be applied to evaluate the goodness of cluster schemes for varying number of clusters (different granularity); a good cluster algorithm that can produce an optimal solution for a fixed number of clusters. To satisfy the first condition, a new validity measure called granularity–dissimilarity (GD) measure is proposed, which is stable in evaluating granularities and works well even when the number of clusters is very large. For the second condition, we propose a new algorithm called multi-step maxmin and merging algorithm (3M algorithm). Experiments show that, when used in conjunction with the new cluster validity measure, 3M algorithm produces better results on the experimental data sets than several alternatives. [Copyright &y& Elsevier]

The simplicial contagion model is employed to study the spreads of two epidemics with mutation in high-order networks. The original epidemic can give birth to a mutated epidemic, but not vice versa. Numerical simulations and mean-field theory results reveal that the spread of the mutated epidemic is entirely dependent on the original epidemic if it cannot spread independently. Conversely, the spread of the original epidemic is entirely inhibited when mutated epidemic spreads by itself. The stability analysis of mean-field theory explains the extinction of the original epidemic and the emergence of a bistable region. Two stable equilibrium points remain unchanged despite variations in parameters like the original epidemic's infection probabilities and mutation rates. While the neighborhood of the stable equilibrium points is regulated by the above parameters. Our conclusions have also been validated in real-world networks. • Two epidemics with mutation spreads in high-order network is investigated. • Mutated epidemic is entirely on the original epidemic if it cannot spread independently. • The spread of the original epidemic is inhibited if the mutated epidemic spreads by itself. • There is a bistable region when mutated epidemics spread independently. [ABSTRACT FROM AUTHOR]

Click-through rate prediction is an important method for online advertising and marketing evaluations. However, for environmental reasons, there is a scarcity and imbalance in the advertising data available. We found that a feature transfer can be applied in a transfer learning method to obtain potential connections from less relevant advertisement data. Considering the complexity and diverse features of advertisement data, a feature transfer cannot allow researchers to discover the relationships among such features within the advertisement data. Therefore, we propose a click-through rate method based on a multi-view feature transfer (MFT). MFT divides the data into common and selected features during the data pretreatment process. It then combines the important feature vectors obtained using a Laplacian matrix with the common features obtained during the pretreatment process to form groups of views. Therefore, combining feature transfer matrix with mutli-view clustering is an innovation of the CTR data prediction process. In our experiments, the MFT model achieved good results. Experiments on a large number of datasets of different sizes and the application of three evaluation indicators show that the MFT method delivers excellent prediction results using the transfer relationships among the characteristics of an advertising dataset, and its performance is better than that of many other advertising click-through rate prediction methods. [ABSTRACT FROM AUTHOR]

Community detection is a challenging issue because most existing methods are not well suited for complex social networks with ambiguous structures. In this paper, we propose a novel community detection method named Stacked Autoencoder-Based Community Detection Method via Ensemble Clustering (CDMEC). This is the first time that we have attempted to apply four different complex network similarity representations to the community detection problem. This work makes up for the insufficiency of the single similarity matrix to describe the similarity relationship between nodes. These similarity representations can fully describe and consider the sufficient local information between nodes in a network topology. Our CDMEC framework combines transfer learning and a stacked autoencoder to obtain an efficient low-dimensional feature representation of complex networks and aggregates multiple inputs through a novel ensemble clustering framework. This novel framework first uses the basic clustering results to construct a consistent matrix, and then it employs the nonnegative matrix factorization (NMF)-based clustering method to detect reliable clustering results from the consistent matrix. The results of various extensive experiments on artificial benchmark networks and real-world networks showed that the proposed CDMEC framework is superior to the existing state-of-the-art community detection methods and has great potential in solving the community detection problems. [ABSTRACT FROM AUTHOR]