Search

The integration of Spiking Neural Networks (Maass, 1997) and Transformers (Vaswani et al., 2017) has significantly enhanced performance in the field, achieving substantial improvements while ensuring energy efficiency. This amalgamation bears significant exploratory significance. This paper aims to enhance the network's capacity for extracting temporal information from neuromorphic datasets. By leveraging the Spikformer (Zhou et al., 2023b) architecture, we introduce a novel network named TE-Spikformer. Through thorough analysis, we identified an imbalance in the Average Firing Rates(AFR) of neurons in the layer preceding the classification head across temporal steps. We thoroughly investigate the root cause of this issue, attributing it to the limitations of the network's use of Batch Normalization (BN) (Ioffe and Szegedy, 2015) layer. To address these challenges, we propose a Batch Group Normalization (BGN) layer. While maintaining the stability of temporal characteristics in the data, we also introduce the Spike Spatio-Temporal Attention(SSTA) module to enhance the network's ability to capture temporal information. To validate the effectiveness of our approach, we conducted multiple experiments using neuromorphic datasets, including DVS-CIFAR10, DVS128 Gesture, and N-Caltech101. The experimental results demonstrate that our algorithm consistently outperforms baseline methods, achieving accuracy rates of 99.30%, 89.60%, and 87.42%, respectively, thereby attaining state-of-the-art performance in the field. • Proposed imbalanced average firing rate across time steps. • Analyzed BN layer inadequacy, proposed BGN layer. • Introduced Spike Spatio-Temporal Attention (SSTA) module. • Achieved SOTA results on three popular neuromorphic datasets. [ABSTRACT FROM AUTHOR]

In this paper, a novel control scheme is developed to solve an optimal containment control problem of unknown continuous-time multi-agent systems. Different from traditional adaptive dynamic programming (ADP) algorithms, this paper proposes an internal reinforcement ADP algorithm (IR-ADP), in which the internal reinforcement signals are added in order to facilitate the learning process. Then a distributed containment control law is designed for each agent with the internal reinforcement signal. The convergence of this IR-ADP algorithm and the stability of the closed-loop multi-agent system are analyzed theoretically. For the implementation of the optimal controllers, three neural networks (NNs), namely internal reinforcement NNs, critic NNs and actor NNs, are utilized to approximate the internal reinforcement signals, the performance indices and optimal control laws, respectively. Finally, some simulation results are provided to demonstrate the effectiveness of the proposed algorithm. [ABSTRACT FROM AUTHOR]

• We are the first to propose QSNN: a quantum algorithm to speed up the classic spiking neural network. • We give the proof for minimum success probability and calculation accuracy of QSNN. • We verify the complexity of QSNN is log-polynomial, and better than the complexity of SNN. • We validate the feasibility and robustness of QSNN on real-world datasets. Spiking neural networks (SNNs) are a kind of neuromorphic computing which meticulously imitates the operations of biological nervous systems. Thus, the SNN is seen as a promising approach to further improve the level of intelligence that the legacy artificial neural networks achieved. In a spiking neuron of an SNN, it is a key step with the highest computing complexity to find out the moment when the output stimuli occur. Classic computer (i.e., electrical computer) based acceleration cannot reduce the computational complexity of the operations in the spiking neurons. As a result, the SNN suffers from scaling and deeper emulation problems. Considering these, in this paper, we propose a quantum algorithm to reduce the complexity of the key steps in the SNN and use this algorithm to build a quantum spiking neuron network (QSNN). More specifically, first, we give mathematical proof that the problem of finding the output stimuli is approximately equal to the problem of calculating the unsigned vector inner products, which can transfer to a quantum operation. Second, we design a scalable quantum circuit of QSNN for data of any dimension and evaluate its basic success probability and calculation accuracy. Third, to improve the QSNN performance, we propose a method to improve the minimum success probability to 99.8%, by repeatedly performing the quantum circuit only 11 times. Fourth, we prove that the computational complexity of the QSNN is a log-polynomial relationship with the data dimension, which is much lower than that of the linear complexity of the classic SNN. Finally, we apply the QSNN to solve classification tasks on two real-world datasets, i.e., MNIST and fashion MNIST which both have two noise levels (i.e., no noise and 50% of the noise). The experiment results show that in addition to the acceleration, QSNN and SNN have equal classification accuracy, which suggests the feasibility and robustness of the QSNN. [ABSTRACT FROM AUTHOR]

Deep neural networks (DNNs) have achieved considerable success in a variety of tasks, and this success has led to an increase in the importance of the security and robustness of algorithms where DNNs are applied. In fact, previous studies have identified that neural networks are vulnerable to adversarial examples that are generated by adding a small amount of noise to the input images. Detection-based methods that can detect adversarial examples among input images are a popular defense technique against adversarial examples. However, these methods reject the detected adversarial images, making it difficult to use the input image for further analysis. In this study, we propose a novel denoising-based detection method that simultaneously denoises and detects an input image when defending against adversarial examples. Our method primarily analyzes images in the feature maps of the target classifier layers based on the consistency between the Fourier domains. We first generate adversarial examples and train the denoising network by minimizing the similarity between the non-adversarial and denoised adversarial images in the feature maps of the Fourier-transformed layers. The predicted denoised images are then input to the adversarial example detector, which determines whether the images are adversarial by exploring the characteristics of the concatenated input and denoised images. There are three main contributions in this paper: (1) We propose a novel denoising-based detection method that simultaneously denoises and detects an input image when defending against adversarial examples. (2) We improve the denoising method for adversarial examples that reconstructs adversarial examples based on the Fourier-transformed feature maps. (3) We improve the detection method by using denoised images in addition to the input images. We tested our proposed method against the fast gradient sign method, basic iterative method, projected gradient descent, Deepfool, and Carlini and Wagner adversarial attack algorithms on two Canadian Institute for Advanced Research (CIFAR) datasets, i.e., CIFAR-10 and CIFAR-100. Our method, which employed a frequency-based denoising mechanism in combination with a detection mechanism, demonstrated significant accuracy improvements in the denoising and detection performances when compared to other state-of-the-art methods. • We propose a method that simultaneously denoises and detects adversarial images. • We enhance denoising by using Fourier-transformed feature maps to reconstruct images. • We improve detection by using both denoised and original input images. [ABSTRACT FROM AUTHOR]

Matrix completion is extensively studied due to its wide applications in science and technology. In this paper, we concentrate our study on the matrix completion problem for image analysis tasks due to their immense importance and pervasive use in many fields. A rich collection of models has been proposed to capture both linear and nonlinear relationships latent in a matrix. Even though nonlinear models possess more powerful matrix completion capabilities than their linear counterparts, these models generally carry higher model complexity and tuning difficulties. To take advantage of the superior discriminative power offered by nonlinear models while curbing its deployment overheads, this paper proposes a novel nonlinear matrix completion model utilizing a deep learning-based approach. In contrast to existing nonlinear models, the new model carefully explores and exploits spatial locality among adjacent matrix elements exhibited in patches of various sizes and locations in a target matrix. Building upon this idea, a new patch-based nonlinear matrix completion algorithm is designed. The algorithm leverages a convolutional neural network to learn the predictive relationship between a matrix element and its surrounding elements through an end-to-end trainable fashion, leading to a capable and easy-to-deploy nonlinear matrix completion solution. To identify an optimal patch size suited for tackling a given matrix completion task without exhaustively enumerating all candidate patch sizes, the new algorithm is coupled with a fast stochastic search procedure, yielding a good trade-off between computational efficiency and accuracy. Extensive experiments are conducted to validate the effectiveness and advantages of the proposed algorithm for nonlinear matrix completion problem in comparison with a series of state-of-the-art algorithms. Experimental results consistently demonstrate the superiority of the new algorithm in completing images with a variety of random and textual noises. [ABSTRACT FROM AUTHOR]

Gradient-based algorithms are commonly used for training radial basis function neural network (RBFNN). However, it is still difficult to avoid vanishing gradient to improve the learning performance in the training process. For this reason, in this paper, an accelerated second-order learning (ASOL) algorithm is developed to train RBFNN. First, an adaptive expansion and pruning mechanism (AEPM) of gradient space, based on the integrity and orthogonality of hidden neurons, is designed. Then, the effective gradient information is constantly added to gradient space and the redundant gradient information is eliminated from gradient space. Second, with AEPM, the neurons are generated or pruned accordingly. In this way, a self-organizing RBFNN (SORBFNN) which reduces the structure complexity and improves the generalization ability is obtained. Then, the structure and parameters in the learning process can be optimized by the proposed ASOL-based SORBFNN (ASOL-SORBFNN). Third, some theoretical analyses including the efficiency of the proposed AEPM on avoiding the vanishing gradient and the stability of SORBFNN in the process of structural adjustment are given, then the successful application of the proposed ASOL-SORBFNN is guaranteed. Finally, to illustrate the advantages of the proposed ASOL-SORBFNN, several experimental studies are examined. By comparing with other existing approaches, the results show that ASOL-SORBFNN performs well in terms of both learning speed and prediction accuracy. [ABSTRACT FROM AUTHOR]

[Display omitted] • A novel polar representation is introduced for video object segmentation. • An asymmetric Siamese network is developed using similar backbones with different depths. • We design the repeated cross correlation and semi-FPN based on peeling convolutions. Video object segmentation (VOS) is an essential part of autonomous vehicle navigation. Besides the accuracy metric, the real-time speed is very important for the algorithms of autonomous vehicles. In this paper, we propose a semi-supervised real-time method based on the Siamese network using a new polar representation. The input of bounding boxes are initialized rather than the object masks, which are applied to the video object detection tasks. The polar representation could reduce the parameters for encoding masks with subtle accuracy loss, so that the algorithm speed can be improved significantly. An asymmetric siamese network is also developed to extract the features from different spatial scales. Moreover, the peeling convolution is proposed to reduce the antagonism among the branches of the polar head. The repeated cross-correlation and semi-FPN are designed based on this idea. The experimental results on the DAVIS-2016 dataset and other public datasets demonstrate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

This paper proposes a change detection algorithm based on Fully Convolutional Siamese Networks for optical aerial images, which is trained by using an improved Contrastive Loss-Focal Contrastive Loss (FCL). The proposed framework equipped with contrastive loss can extract features directly from image pairs and measure changes by using a distance metric. In other words, this method encourages reducing intra-class variance and enlarging inter-class difference, so that the binarized change map can be obtained by a simple threshold. In change detection task, a critical problem is how to overcome example imbalance (i.e. unchanged examples are much more than changed examples). To address this challenge, a novel focal contrastive loss is proposed to further improve the performance of the model. FCL can reduce the impact of example imbalance and make the model focus learning on hard examples. Extensive experiments demonstrate that the proposed approach is more abstract as well as robust. Compared with other baseline methods, the presented method achieves better results on SAZDA, TISZADOB, CDD, and WHU-CD data set. It achieves state-of-the-art performance in terms of the F1 measure. [ABSTRACT FROM AUTHOR]

Gradient-based algorithms are commonly used for training radial basis function neural network (RBFNN). However, one of the challenges in the training process is determining how to avoid vanishing gradient. To solve this problem, an accelerated gradient algorithm (AGA) is designed to improve the learning performance of RBFNN in this paper. First, an indirect detection mechanism, based on the instantaneous gradient decay rate (IGDR) and instantaneous convergence rate (ICR), is developed to identify the vanishing gradient in learning process. Second, an amplification gradient strategy (AGS), which can increase the gradient value of learning parameters, is designed to accelerate the learning speed of RBFNN. Third, the analysis of AGA-based RBFNN (AGA-RBFNN) is given to guarantee the successful application. Finally, some benchmark and real problems are used to illustrate the effectiveness of AGA-RBFNN. The results demonstrate the effectiveness of AGA-RBFNN. [ABSTRACT FROM AUTHOR]

An algorithm that generates an exponential number of stable states for the very well-known Hopfield Neural Network (HNN) is introduced in this paper. We show that the quantity of stable states depends on the dimension and number of components of the input pattern supporting noise. Extensive tests verify that the states generated by our algorithm are stable states and show the exponential storage capacity of a HNN. This paper opens the possibility of designing improved HNNs able to achieve exponential storage, and thus find their applicability in complex real-world problems. [ABSTRACT FROM AUTHOR]

In this paper, a hierarchical method for human body segmentation in static images is proposed. Based on the face detection, a torso model is developed to locate the torso to provide the foreground seeds for torso segmentation with graph cuts. Similarly, an upper leg model is designed for lower body segmentation based on the segmented torso. Besides, a general method which combines shape information into graph cuts is presented, which can make the segmentation more accurate. The main contributions of this paper include two models and a new method of integrating shape information into graph cuts. Experiments on our collected real world images show that our method can accurately recover human body from static images with a variety of individuals, poses, backgrounds and clothing. [ABSTRACT FROM AUTHOR]

The problem of subspace clustering which refers to segmenting a collection of data samples approximately drawn from a union of linear subspaces is considered in this paper. Among existing subspace clustering algorithms, low rank representation (LRR) based subspace clustering is a very powerful method and has demonstrated that its performance is good. Latent low rank representation (LLRR) subspace clustering algorithm is an improvement of the original LRR algorithm when the observed data samples are insufficient. The clustering accuracy of LLRR is higher than that of LRR. Recently, Frobenius norm minimization based LRR algorithm has been proposed and its clustering accuracy is higher than that of LRR demonstrating the effectiveness of Frobenius norm as another convex surrogate of the rank function. Combining LLRR and Frobenius norm, a new low rank representation subspace clustering algorithm is proposed in this paper. The nuclear norm in the LLRR algorithm is replaced by the Frobenius norm. The resulting optimization problem is solved via alternating direction method of multipliers (ADMM). Experimental results show that compared with LRR, LLRR and several other state-of-the-art subspace clustering algorithms, the proposed algorithm can get higher clustering accuracy. Compared with LLRR, the running time of the proposed algorithm is reduced significantly. [ABSTRACT FROM AUTHOR]

In this paper, we propose a new neural network-based iterative learning leader-less consensus control algorithm for the second-order uncertain time-varying nonlinear multi-agent systems, where the position of each agent is constrained in an open interval which can be pre-selected and the control direction can be non-identical and partially unknown. With the help of a novel barrier composite energy function and multiple piece-wise Nussbaum functions, the designed completely distributed learning protocols can make the positions and velocities of the agents achieve the exact consensus on a finite time interval as the iteration number increases to infinity. Meanwhile, the position trajectories of the agents at each iteration do not exceed the constraints bounds. Besides, to testify the feasibility and utility of the presented algorithm, two simulation examples are given in the end. [ABSTRACT FROM AUTHOR]

Deep learning is a branch of neural network which has been intensively developed in the last decade. Due to the high-accuracy classification ability, the deep learning algorithms have been widely used in many fields, such as speech recognition, image recognition, and natural speech processing. However, they also show some shortcomings especially on the selection of some parameters in the network, including hyper-parameters, which is still treated as a time consuming task. In this paper, a modified ABC (ABC-ISB) optimization algorithm is proposed to automatically train the parameters of Feed-Forward Artificial Neural Networks, which is a typical a neural network. In the proposed ABC algorithm, we utilize the information of neighbors with better performance to accelerate the convergence of employed and onlooker bees respectively. In addition, a new selection strategy and a gbest-guided strategy are introduced to enhance the global search capability and balance the exploration and exploitation of the algorithm separately. The experimental results show our ABC-ISB is generally leading and competitive. [ABSTRACT FROM AUTHOR]

Neural network learning is usually time-consuming since backpropagation needs to compute full gradients and backpropagate them across multiple layers. Despite its success of existing works in accelerating propagation through sparseness, the relevant theoretical characteristics remain under-researched and empirical studies found that they suffer from the loss of information contained in unpropagated gradients. To tackle these problems, this paper presents a unified sparse backpropagation framework and provides a detailed analysis of its theoretical characteristics. Analysis reveals that when applied to a multilayer perceptron, our framework essentially performs gradient descent using an estimated gradient similar enough to the true gradient, resulting in convergence in probability under certain conditions. Furthermore, a simple yet effective algorithm named m emorized s parse b ack p ropagation (MSBP) is proposed to remedy the problem of information loss by storing unpropagated gradients in memory for learning in the next steps. Experimental results demonstrate that the proposed MSBP is effective to alleviate the information loss in traditional sparse backpropagation while achieving comparable acceleration. [ABSTRACT FROM AUTHOR]

• Unsupervised framework for depth estimation and camera motion prediction. • Depth CNN and pose CNN are trained jointly and can be used respectively. • The supervision signal with a left-right consistency is constructed by spatial and temporal geometry constraints. • Results outperform previous unsupervised methods and some supervised methods. • A model which is trained on the Euroc dataset is used to test the algorithm's generalization capability. Depth estimation from monocular video plays a crucial role in scene perception. The significant drawback of supervised learning models is the need for vast amounts of manually labeled data (ground truth) for training. To overcome this limitation, unsupervised learning strategies without the requirement for ground truth have achieved extensive attention from researchers in the past few years. This paper presents a novel unsupervised framework for estimating single-view depth and predicting camera motion jointly. Stereo image sequences are used to train the model while monocular images are required for inference. The presented framework is composed of two CNNs (depth CNN and pose CNN) which are trained concurrently and tested independently. The objective function is constructed on the basis of the epipolar geometry constraints between stereo image sequences. To improve the accuracy of the model, a left-right consistency loss is added to the objective function. The use of stereo image sequences enables us to utilize both spatial information between stereo images and temporal photometric warp error from image sequences. Experimental results on the KITTI and Cityscapes datasets show that our model not only outperforms prior unsupervised approaches but also achieving better results comparable with several supervised methods. Moreover, we also train our model on the Euroc dataset which is captured in an indoor environment. Experiments in indoor and outdoor scenes are conducted to test the generalization capability of the model. [ABSTRACT FROM AUTHOR]

Spiking neural networks (SNNs) have attracted widespread attention due to their unique bio-interpretability and low-power properties, but the non-differentiability of discrete spike sequences fired by spiking neurons brings difficulties to the learning of SNNs. Recently, surrogate gradient (SG) and back-propagation through time (BPTT) have provided an excellent idea for training SNNs constructed by leaky integrate-and-fire (LIF) neuron models. The LIF neuron model has been widely used in previous SNNs due to its simplicity and low computational cost, however, this also limits its simulation of biological neuron dynamics, reducing the biological interpretability of SNNs. In this paper, we generalize SG and BPTT to SNNs constructed by spike response model (SRM) and propose the BP-SRM algorithm. Specifically, we address why BPTT succeeded in LIF neurons-based SNNs but failed in SRM neurons-based SNNs in previous researches. Then we establish an iterative form of the SRM neuron model by selecting different state variables. Based on iterable SRM, we get the spatiotemporal dependencies between the state variables of SRM neurons-based SNNs, which allows us to derive the gradient and update weights in SNN by BP-SRM. Then we design temporal channel normalization for BP-SRM and verify the performance of the SNNs on static image dataset, dynamic image dataset and engineering dataset, including Fashion-MNIST, Weather Dataset, N-MNIST, American Sign Language Dataset and bearing fault diagnosis dataset. The experiment results indicate that BP-SRM achieves the state-of-the-art performance of SNNs. [ABSTRACT FROM AUTHOR]

This survey makes an overview of the most recent applications on the neural networks for the computer-aided medical diagnosis (CAMD) over the past decade. CAMD can facilitate the automation of decision making, extraction and visualization of complex characteristics for clinical diagnosis purposes. Over the past decade, neural networks have attained considerable research interest and are widely employed to complex CAMD systems in diverse clinical application domains, such as detecting diseases, classification of diseases, testing the compatibility of new drugs, etc. Overall, this paper reviews the state-of-the-art of neural networks for CAMD. It helps the readers understand the topic of neural networks for CAMD by summarizing the findings addressed in recent academic papers as well as presenting a few open issues of developing the research on this topic. [ABSTRACT FROM AUTHOR]

• We propose a new method for short-term load forecasting (STLF). • The new method is based on artificial neural networks optimized using bat algorithm. • The method has the capability to "shape and fine-tune" the parameters. • We tested the proposed method in Portuguese, New England (USA) and New York city load time-series. • The proposed approach exhibits a better forecasting accuracy. Short-term load forecasting is very important for reliable power system operation, even more so under electricity market deregulation and integration of renewable resources framework. This paper presents a new enhanced method for one day ahead load forecast, combing improved data selection and features extraction techniques (similar/recent day-based selection, correlation and wavelet analysis), which brings more "regularity" to the load time-series, an important precondition for the successful application of neural networks. A combination of Bat and Scaled Conjugate Gradient Algorithms is proposed to improve neural network learning capability. Another feature is the method's capacity to fine-tune neural network architecture and wavelet decomposition, for which there is no optimal paradigm. Numerical testing using the Portuguese national system load, and the regional (state) loads of New England and New York, revealed promising forecasting results in comparison with other state-of-the-art methods, therefore proving the effectiveness of the assembled methodology. [ABSTRACT FROM AUTHOR]

Abstract Uncertainty is a fundamental and unavoidable feature in daily life, which is the same for a single classifier. Thus, combining the predictions of many different classifiers is a very successful way to reduce the uncertainty. In this paper, we present a Correcting Reliability Level (CRL) supervised three-way decision (3WD) cascade model to implement image classification tasks. Our model simulates the human decision process by using 3WD to judge "certainty" or "uncertainty" of the classification result. When judged as "uncertainty", CRL will supervise the 3WD and learn more information to make the final prediction. In addition, we introduce two Class Grouping methods to mining the relation between labels, which help us to train several expert ConvNets for different types of images. Experimental results show that our model can effectively reduce the classification error rate compared with the base classifier. [ABSTRACT FROM AUTHOR]

Abstract A Deep Boltzmann Machine is a model of a Deep Neural Network formed from multiple layers of neurons with nonlinear activation functions. The structure of a Deep Boltzmann Machine enables it to learn very complex relationships between features and facilitates advanced performance in learning of high-level representation of features, compared to conventional Artificial Neural Networks. Feature selection at the input level of Deep Neural Networks has not been well studied, despite its importance in reducing the input features processed by the deep learning model, which facilitates understanding of the data. This paper proposes a novel algorithm, Deep Feature Selection (Deep-FS), which is capable of removing irrelevant features from large datasets in order to reduce the number of inputs which are modelled during the learning process. The proposed Deep-FS algorithm utilizes a Deep Boltzmann Machine, and uses knowledge which is acquired during training to remove features at the beginning of the learning process. Reducing inputs is important because it prevents the network from learning the associations between the irrelevant features which negatively impact on the acquired knowledge of the network about the overall distribution of the data. The Deep-FS method embeds feature selection in a Restricted Boltzmann Machine which is used for training a Deep Boltzmann Machine. The generative property of the Restricted Boltzmann Machine is used to reconstruct eliminated features and calculate reconstructed errors, in order to evaluate the impact of eliminating features. The performance of the proposed approach was evaluated with experiments conducted using the MNIST, MIR-Flickr, GISETTE, MADELON and PANCAN datasets. The results revealed that the proposed Deep-FS method enables improved feature selection without loss of accuracy on the MIR-Flickr dataset, where Deep-FS reduced the number of input features by removing 775 features without reduction in performance. With regards to the MNIST dataset, Deep-FS reduced the number of input features by more than 45%; it reduced the network error from 0.97% to 0.90%, and also reduced processing and classification time by more than 5.5%. Additionally, when compared to classical feature selection methods, Deep-FS returned higher accuracy. The experimental results on GISETTE, MADELON and PANCAN showed that Deep-FS reduced 81%, 57% and 77% of the number of input features, respectively. Moreover, the proposed feature selection method reduced the classifier training time by 82%, 70% and 85% on GISETTE, MADELON and PANCAN datasets, respectively. Experiments with various datasets, comprising a large number of features and samples, revealed that the proposed Deep-FS algorithm overcomes the main limitations of classical feature selection algorithms. More specifically, most classical methods require, as a prerequisite, a pre-specified number of features to retain, however in Deep-FS this number is identified automatically. Deep-FS performs the feature selection task faster than classical feature selection algorithms which makes it suitable for deep learning tasks. In addition, Deep-FS is suitable for finding features in large and big datasets which are normally stored in data batches for faster and more efficient processing. [ABSTRACT FROM AUTHOR]

Abstract The purpose of image fusion is to integrate useful information from multiple images and produce a more reliable image. The key problem of multi-focus image fusion is how to determine the focused regions of the source images. As an effective and excellent fusion algorithm, the focused regions in the source images should be preserved as much as possible into the fused image. To accomplish this goal, a novel multi-focus image fusion method based on a focused regions boundary finding and multi-scale transform (MST) is proposed in this paper. The Meanshift algorithm is used to determine the focused regions first. Then, an edge detection method and morphological method are used to find the boundaries of the focused regions in the source images. For the focused boundary regions, the combination of pulse coupled neural network (PCNN) and Gaussian fuzzy method is used to produce the fused boundary region in nonsubsampled contourlet transform (NSCT) domain. Finally, the fused boundary region and the focused region of the source images are fused directly. The experimental results demonstrate that the proposed algorithm can accurately determine the focused regions, and at the same time, a better fused boundary region can be obtained; this algorithm is superior to conventional methods with respect to both objective quality evaluations and visual inspection. [ABSTRACT FROM AUTHOR]

The geometric measure of entanglement of a multipartite pure state is defined it terms of its geometric distance from the set of separable pure states. The quantum eigenvalue problem is derived to compute the separable pure state nearest to the given multipartite pure state. Computing the modulus largest quantum eigenvalue for a multipartite pure state is equivalent to finding the best complex rank-one approximation of the complex unit tensors, associated with the multipartite pure states. This paper is devoted to present a complex-valued neural networks approach for the computation of the quantum eigenvalue problem for multipartite pure states. We design the neural networks for computing the best rank-one tensor approximation of complex tensors, and prove that the solution of the networks is locally asymptotically stable in the sense of Lyapunov stability theory. This solution also converges to the local optimal solutions of the best complex rank-one tensor approximation. We illustrate our theoretical results via numerical simulations. [ABSTRACT FROM AUTHOR]

Correntropy is a local similarity measure between two arbitrary variables, and it has been applied in a variety of learning algorithms to improve noise insensitivity. In this paper, based on the correntropy, a non-convex and bounded loss function is obtained which contains second and higher order moments of the classification margin. And the novel loss function is robust to noises and close to the 0–1 loss function. Then we introduce it into extreme learning machine (ELM), and propose a correntropy-based robust ELM framework for classification, trained by half quadratic optimization to cope with non-convexity of the algorithm. To evaluate robustness, feature noise and label noise are simulated to provide noisy environments. Experimental results on benchmark datasets demonstrate that the proposed algorithm is better than original algorithms and robust algorithms. Moreover, the superiority of proposed algorithm in noisy environment is more evident, which further proves its robustness to noises. [ABSTRACT FROM AUTHOR]

This paper mainly focuses on how to construct concept lattice effectively and efficiently based on improved variable precision rough set. On the basis of preprocessing formal concept, one algorithm that can determine the value range of variable precision parameter β according to the approximate classification quality is proposed. An improved β -upper and lower distribution attribute reduction algorithm is also proposed based on the improved variable precision rough set, the algorithm can be used for attribute reduction on the original data of the concept lattice, and to eliminate the redundant knowledge or noises of the formal context. For the reduced formal context, the paper combines the concept construction algorithm with an improved rule acquisition algorithm seamlessly, and proposes a novel approach of concept lattice construction based on improved variable precision rough set. Finally, a concept lattice generation prototype system is developed, this paper also performs comprehensive experiments, and the effectiveness of the improved algorithm is proved through the experimental results. [ABSTRACT FROM AUTHOR]

Deep learning algorithms are a subset of the machine learning algorithms, which aim at discovering multiple levels of distributed representations. Recently, numerous deep learning algorithms have been proposed to solve traditional artificial intelligence problems. This work aims to review the state-of-the-art in deep learning algorithms in computer vision by highlighting the contributions and challenges from over 210 recent research papers. It first gives an overview of various deep learning approaches and their recent developments, and then briefly describes their applications in diverse vision tasks, such as image classification, object detection, image retrieval, semantic segmentation and human pose estimation. Finally, the paper summarizes the future trends and challenges in designing and training deep neural networks. [ABSTRACT FROM AUTHOR]

Visual recognition of crowd dynamics has had a huge impact on several applications including surveillance, situation awareness, homeland security and intelligent environments. However, the state-of-the-art in crowd analysis has become diverse due to factors such as: (a) the underlying definition of a crowd, (b) the constituent elements of the crowd processing model, (c) its application, hence (d) the dataset and (e) the evaluation criteria available for benchmarking. Although such diversity is healthy, the techniques for crowd modelling thus developed have failed to establish credibility therefore becoming unreliable and of questionable validity across different research disciplines. This paper aims to give an account of such issues by deducing key statistical evidence from the existing literature and providing recommendations towards focusing on the general aspects of techniques rather than any specific algorithm. The advances and trends in crowd analysis are drawn in the context of crowd modelling studies published in leading journals and conferences over the past 7 years. Finally, this paper shall also provide a qualitative and quantitative comparison of some existing methods using various publicly available crowd datasets to substantiate some of the theoretical claims. [ABSTRACT FROM AUTHOR]

Twin support vector machines (TWSVM) is a new machine learning method based on the theory of Support Vector Machine (SVM). Unlike SVM, TWSVM would generate two non-parallel planes, such that each plane is closer to one of the two classes and is as far as possible from the other. In TWSVM, a pair of smaller sized quadratic programming problems (QPPs) is solved, instead of solving a single large one in SVM, making the computational speed of TWSVM approximately 4 times faster than the standard SVM. At present, TWSVM has become one of the popular methods because of its excellent learning performance. In this paper, the research progress of TWSVM is reviewed. Firstly, it analyzes the basic theory of TWSVM, then tracking describes the research progress of TWSVM including the learning model and specific applications in recent years, finally points out the research and development prospects. This helps researchers to effectively use TWSVM as an emerging research approach, encouraging them to work further on performance improvement. [ABSTRACT FROM AUTHOR]

In this paper, we propose a semi-supervised facial landmark detection algorithm (SEMI) based on convolutional neural network (CNN), which can detect facial components and landmarks simultaneously. Unlike previous coarse-to-fine algorithms, our model does not need extra input such as initial landmark prediction. It also solves the occlusion problem of large area by detecting the visible facial components while existing face detectors failed to detect faces. Semi-supervised learning algorithm is also an effective data augmentation method. In our experiment, each image has two types of ground truth, one is bounding-box related (classification and coordinates) and the other is landmark coordinates inside the bounding-box. The supervised data have both two types of ground truth while the semi-supervised data only have the bounding-box. Our model was trained by the merge of two parts of data. Extensive evaluations on Helen, LFPW and 300-W show that our algorithm is able to complete the landmark task and performs better than many state-of-the-art facial landmark detecting algorithms. [ABSTRACT FROM AUTHOR]

This paper presents a distributed scheme for the control of multiple redundant manipulators to simultaneously achieve four objectives, i.e., the task to reach global cooperation, joint-physical limits compliance, limited communications among manipulators and optimality in terms of a specified performance index. In addition, corresponding theoretical analyses are provided, which guarantee that, with the communication network being connected, all manipulators can jointly obtain the same desired motion information. Then, the proposed scheme is converted into a quadratic program (QP) formed formulation and solved by a dynamic neural network with rigorously provable convergence. Furthermore, simulations and comparisons are provided to illustrate the effectiveness of the proposed distributed scheme as well as the presented dynamic neural network. [ABSTRACT FROM AUTHOR]

Elman neural network (ENN) is one of recurrent neural networks (RNNs). Comparing to traditional neural networks, ENN has additional inputs from the hidden layer, which forms a new layer–the context layer. So the standard back-propagation (BP) algorithm used in ENN is called Elman back-propagation algorithm (EBP). ENN can be applied to solve prediction problems of discrete time sequence. However, the EBP algorithm suffers from low convergence speed and poor generalization performance. To solve this problem, a new learning rate scheme is proposed, the convergence of new proposed scheme is proved. Furthermore, the contrast experiment is utilized to demonstrate the effectiveness of the proposed scheme from the aspects of convergence speed and consumption time with some popular schemes such as the original ENN, and PSO–ENN which uses PSO algorithm to search the best structure of ENN. The experience shows that the modified method proposed in this paper works best. [ABSTRACT FROM AUTHOR]

Nonnegative matrix factorization (NMF) is a very popular unsupervised or semi-supervised learning method useful in various applications including data clustering, image processing, and semantic analysis of documents. This study focuses on Symmetric NMF (SNMF), which is a special case of NMF and can be useful in network analysis. Although there exist several algorithms for SNMF in literature, their convergence and initialization have not been well addressed. In this paper, we first discuss the convergence and initialization of existing algorithms for SNMF. We then propose a Converged Algorithm for SNMF (called CASNMF) which minimizes the Euclidean distance between a symmetrical matrix and its approximation of SNMF. Based on the optimization principle and the local auxiliary function method, we prove that our presented CASNMF does not only converge to a stationary point, but also could be applied to the wider range of SNMF problems. In addition, CASNMF does not require that the initial values are nonzero. To verify our theoretical results, experiments on three data sets are conducted by comparing our proposed CASNMF with other existing methods. [ABSTRACT FROM AUTHOR]

There are two categories of partial shape matching problem: whole-to-part matching and part-to-part matching. The first establishes the best match between an open curve and a part of a closed curve, while the second is to find a part of the first input closed curve that can be well aligned with a part of the second input closed curve. In this paper, we present a novel approach to solve these two categories of the partial shape matching problem. First, we propose a novel shape descriptor, triangular centroid distances (TCDs), for shape representation; the TCDs shape descriptor is invariant to translation, rotation, scaling, and considerable shape deformations. Then, using the TCDs shape descriptor, we present a method to deal with the whole-to-part partial shape matching problem. Finally, we extend our work to part-to-part partial shape matching. Here, we propose a new approach, again using the TCDs shape descriptor, to solve the part-to-part partial shape matching problem. Experimental results demonstrate that our method outperforms existing methods in 2D nonrigid partial shape matching. [ABSTRACT FROM AUTHOR]

Word embeddings and CNN (convolutional neural networks) architecture are crucial ingredients of sentiment analysis. However, sentiment and lexicon embeddings are rarely used and CNN is incompetent to capture global features of sentence. To this end, semantic embeddings, sentiment embeddings and lexicon embeddings are applied for texts encoding, and three different attentions including attention vector, LSTM (long short term memory) attention and attentive pooling are integrated with CNN model in this paper. Additionally, a word and its context are explored to disambiguate the meaning of the word for rich input representation. To improve the performance of three different attention CNN models, CCR (cross-modality consistent regression) and transfer learning are presented. It is worth noticing that CCR and transfer learning are used in textual sentiment analysis for the first time. Finally, some experiments on two different datasets demonstrate that the proposed attention CNN models achieve the best or the next-best results against the existing state-of-the-art models. [ABSTRACT FROM AUTHOR]

In this paper, the problem of event-triggered output feedback synchronization control for complex dynamical networks (CDN) is studied. Based on a distributed state observer, a novel event-triggered mechanism is firstly introduced to design the output feedback synchronization controller, and by applying a useful graph theoretical result, the asymptotic convergence of the synchronization errors is then guaranteed. Moreover, it is proved that the designed event-triggered synchronization method is also valid for the CDNs with time-varying coupling weights. In particular, the proposed method is an extension of the existing state feedback event-triggered synchronization results. Finally, the circuit realizations for the new event-triggered mechanism and control approach are given, and the effectiveness of the proposed scheme is illustrated through the simulation results. [ABSTRACT FROM AUTHOR]

In big data fields, with increasing computing capability, artificial neural networks have shown great strength in solving data classification and regression problems. The traditional training of neural networks depends generally on the error back propagation method to iteratively tune all the parameters. When the number of hidden layers increases, this kind of training has many problems such as slow convergence, time consuming, and local minima. To avoid these problems, neural networks with random weights (NNRW) are proposed in which the weights between the hidden layer and input layer are randomly selected and the weights between the output layer and hidden layer are obtained analytically. Researchers have shown that NNRW has much lower training complexity in comparison with the traditional training of feed-forward neural networks. This paper objectively reviews the advantages and disadvantages of NNRW model, tries to reveal the essence of NNRW, gives our comments and remarks on NNRW, and provides some useful guidelines for users to choose a mechanism to train a feed-forward neural network. [ABSTRACT FROM AUTHOR]

This paper presents a novel unsupervised multi-source domain adaptation approach, named as coupled local–global adaptation (CLGA). At the global level, in order to maximize the adaptation ability, CLGA regards multiple domains as a unity, and jointly mitigates the gaps of both marginal and conditional distributions between source and target dataset. At the local level, with the intention of maximizing the discriminative ability, CLGA investigates the relationship among distinctive domains, and exploits both class and domain manifold structures embedded in data samples. We formulate both local and global adaptation in a concise optimization problem, and further derive an analytic solution for the objective function. Extensive evaluations verify that CLGA performs better than several existing methods not only in multi-source adaptation tasks but also in single source scenarios. [ABSTRACT FROM AUTHOR]

In this paper, the dynamics including boundedness and stability for a general class of complex-valued neural networks with variable coefficients and proportional delays are investigated. By employing inequality techniques and mathematical analysis method, some sufficient criteria to guarantee boundedness and global exponential stability are established for the considered neural networks. As a special case that the coefficients of networks are constants, sufficient criteria are also derived to guarantee the existence, uniqueness and global exponential stability of the equilibrium point. This work generalizes and improves previously known results, and the obtained criteria can be tested and applied easily in practice. An illustrative example demonstrates the feasibility of the proposed results. [ABSTRACT FROM AUTHOR]

This paper proposes a novel echo state network (ESN) architecture in a deep learning framework for time series prediction. The architecture is a uniform and consistent system with functional parts of the pre-training input network that effectively captures information with different degrees of abstraction in the observed data, and the minimum complexity ESN that possesses the powerful nonlinear approximation capability and highly efficient training. To our best knowledge, this is the first systematic model attempting to introduce the deep learning methodology to the ESN modeling, which provides a more robust alternative to the conventional shallow ESNs. Extensive experiments on various widely used benchmarks of different origins and features show that our model achieves a great enhancement in the prediction accuracy and short-term memory capacity, without significant tradeoff in the model’s computational efficiency. [ABSTRACT FROM AUTHOR]

In this paper, the optimal performance of multi-input and multi-output (MIMO) linear time-invariant (LTI) discrete-time networked control systems (NCSs) are investigated based on state-space representation. In communication channels, channel fading, additive white Gaussian noise (AWGN) and scaling parameter matrix are considered. An average performance index is constructed. Moreover, the control energy and the output energy of the system are considered simultaneously in the output regulation performance. The scaling parameter matrix is introduced, which may be considered as an encoder or modification on the channel. The optimal regulation performances of the system are obtained, and the relationship is given between the optimal regulation performances and the steady-state solution of the corresponding algebraic Riccati equation (ARE). Results indicate that the optimal performance is influenced by the non-minimum phase zeros, unstable poles, the scaling parameter matrix, and the characteristics of the communication noise and fading. In addition, the channel power constraint is considered in Theorem 2 . We introduce a novel power weighting factor, which can weigh the relative importance of system output and channel input energy. The obtained results will give some guidance for the design of NCSs. Finally, a simulation example is given to demonstrate effectiveness of the theoretical results. [ABSTRACT FROM AUTHOR]

On one hand, the detection of tightly connected groups, also known as community detection in complex networks, is a prominent problem for network analysis and mining. On the other hand, almost all of social, biological, bibliographic, communication and computer systems are modeled as temporal networks, the topological structures of which evolve with time, or multiplex networks, each pair of nodes of which has multiple linked relations. Current methods of community detection for temporal networks are based on incremental, independent or evolutionary clustering, and for multiplex networks are based on fusion of the multiple links. However, all these methods ignore the common structure hidden in the networks, which is denoted as the common cluster here. So in this paper, we propose a constrained common cluster based model ( C 3 model) to analyze the temporal and multiplex networks, which can not only detect the community structure, but also identify the importance of each node based on the common cluster structure of both two classes of networks. The intrinsic assumption of the proposed model is that there are common or coincident clusters hidden in these networks. In detail, we first construct the Markov steady-state matrices of each snapshot of temporal network or each slice of multiplex network. Next, we propose the object function of C 3 model by combining the Markov steady-state matrices, similarity matrices with community membership matrices of each snapshot or slice of the network. Last, a gradient descent algorithm based on non-negative matrix factorization is proposed for the object function. Experiments on both synthetic datasets and real-world networks demonstrate that the proposed C 3 model has competitive performance based on the evaluation indexes NMI and error of community detection, otherwise, the proposed model could identify the importance of nodes of the temporal or multiplex networks. [ABSTRACT FROM AUTHOR]

In this paper, the problem of fault-tolerant control (FTC) is investigated for a class of multivariable stochastic nonlinear systems in the strict-feedback form. The considered systems possess unknown nonlinear functions, unmeasured states, virtual control coefficients and actuator faults (bias and gain faults). Neural networks (NNs) are adopted to approximate the unknown nonlinear functions. Then, a state observer is constructed to solve the problem of unmeasured states. The problem of “explosion of complexity” is solved by using dynamic surface control (DSC) method. Based on the combination of Nussbaum gain function with Lyapunov function stability theory, an adaptive NNs output feedback FTC method is developed in the frame of adaptive backstepping design technique. It is shown that all signals in the closed-loop system are proved to be bounded in probability, and the system output can follow the given reference signal well. The effectiveness of the proposed approach is verified by a simulation example. [ABSTRACT FROM AUTHOR]

In online learning, the contribution of old samples to a model decreases as time passes, and old samples gradually become invalid. Although the Online Sequential Extreme Learning Machine (OS-ELM) can avoid the repetitive training of old samples, invalid samples are still used, which goes against improving the accuracy of an OS-ELM model. The Online Sequence Extreme Learning Machine with Forgetting Mechanism (FOS-ELM) timely discards invalid samples, but it does not consider the differences among valid samples and then has the limitation on boosting the accuracy and generalization. To solve this issue, the Memory Degradation Based OS-ELM (MDOS-ELM) is proposed in this paper. The MDOS-ELM adjusts the weights of the old and new samples in real time by a self-adaptive memory factor, and simultaneously discards invalid samples. The self-adaptive memory factor is determined by two elements. One is the similarity between the new and old samples, and the other is the prediction errors of the current training samples on the previous model. The performance of the proposed MDOS-ELM is validated on both regression and classification datasets which include an artificial dataset and twenty-two real-world dataset. The results demonstrate that the MDOS-ELM model outperforms the OS-ELM and the FOS-ELM models on the accuracy and generalization. [ABSTRACT FROM AUTHOR]

Compressed sensing (CS) has been one of the great successes of applied mathematics in the last decade. This paper proposes a new method, combining the advantage of the Compressive Sampling Matching Pursuit (CoSaMP) algorithm and the Quasi–Newton Iteration Projection (QNIP) algorithm, for the recovery of sparse signal from underdetermined linear systems. To get the new algorithm, Quasi–Newton Projection Pursuit (QNPP), the least-squares technique in CoSaMP is used to accelerate convergence speed and QNIP is modified slightly. The convergence rate of QNPP is studied, under a certain condition on the restricted isometry constant of the measurement matrix, which is smaller than that of QNIP. The fast version of QNPP is also proposed which uses the Richardson’s iteration to reduce computation time. The numerical results show that the proposed algorithms have higher recovery rate and faster convergence speed than existing techniques. [ABSTRACT FROM AUTHOR]

Compared to studies of analyzing hyperspectral image (HSI), finding spectral rectification, especially the seriously distorted pixels in boundaries, was seldom addressed in a clear way, albeit of first importance in HSI analysis and interpretation. In this paper, we present a simple but promising rectification method works in a coarse-to-fine framework for removing noise and enhancing useful features. Our approach, called CSR (coarse-to-fine spectral rectification), combines the theory of scale-aware with local smoothness for HSI rectification problem that is seldom pointed out. The useless information like noise in small scale is removed firstly. Then, the distinctive information like boundary in large scale is enhanced. The experimental result enjoys a built-in smoothing effect and a fact of the identical materials with same or similar signatures, which is suited for HSI subsequent application. Furthermore, our approach has powerful influence on both classification via five classifiers in terms of class labeled data and unmixing regarding to class unlabeled data. Rectified by a coarse-to-fine framework, our method presents superior performance and runs much faster than the competing methods. [ABSTRACT FROM AUTHOR]

Motivated by the recent growing interest in pairwise learning problems, we study the generalization performance of Online Pairwise lEaRning Algorithm (OPERA) in a reproducing kernel Hilbert space (RKHS) without an explicit regularization. The convergence rates established in this paper can be arbitrarily closed to O ( T − 1 2 ) within T iterations and largely improve the existing convergence rates for OPERA. Our novel analysis is conducted by showing an almost boundedness of the iterates encountered in the learning process with high probability after establishing an induction lemma on refining the RKHS norm estimate of the iterates. [ABSTRACT FROM AUTHOR]

One of the most active areas in computer vision is object detection, which has made significant improvement in recent years. Current state-of-the-art object detection methods mostly adhere to the framework of the regions with convolutional neural network (R-CNN). However, they only take advantage of the local appearance features inside object bounding boxes. Since these approaches ignore the contextual information around the object proposals, the outcome of these detectors may generate a semantically incoherent interpretation of the input image. In this paper, we propose a novel object detection system which incorporates the local appearance and the contextual information. Specifically, the contextual information comprises the relationships among objects and the global scene based contextual feature generated by a convolutional neural network. The whole system is formulated as a fully connected conditional random field (CRF) defined on object proposals. Then the contextual constraints among object proposals are modeled as edges naturally. Furthermore, a fast mean field approximation method is utilized to infer in this CRF model efficiently. The experimental results demonstrate that our algorithm achieves a higher mean average precision (mAP) on PASCAL VOC 2007 datasets compared with the baseline algorithm Faster R-CNN. [ABSTRACT FROM AUTHOR]

In this paper, a novel multilayer backpropagation (BP) neural network model is proposed based on conjugate gradient (CG) method with generalized Armijo search. The presented algorithm requires low memory and performs fast convergent speed in practical applications. One reason is that the constructed conjugate direction guarantees the sufficient descent behavior in minimizing the given objective function. The other stems from the fact that the generalized Armijo method can automatically determine a more suitable learning rate in each training epoch. As a theoretical contribution, two deterministic convergent results, weak and strong convergence, have been detailedly proved under more relaxed assumptions. The weak convergence means that the norm of gradient of the objective function tends to zero. For the strong convergence, it represents that the sequence of weight vectors approaches a fixed point. To support the theoretical results, some illustrated simulations have been done on various benchmark datasets. [ABSTRACT FROM AUTHOR]

Clustering analysis has been applied in all aspects of data mining. Density-based and grid-based clustering algorithms are used to form clusters from the core points or dense grids to extend to the boundary of the clusters. However, deficiencies are still existed. To find out the right boundary and improve the precision of the cluster, this paper has proposed a new clustering algorithm (named C-USB) based on the skew characteristic of the data distribution in the cluster margin region. The boundary degree calculated by skew degree and the local density are used to distinguish whether a data is an internal point or non-internal point. And the connected matrix is constructed by removing the neighbor relationships of non-internal points from the relationships of all points, then the clusters can be formed by searching from the connected matrix towards internal of the clusters. Experimental results on synthetic and real data sets show that the C-USB has higher accuracy than that of similar algorithms. [ABSTRACT FROM AUTHOR]

This paper is concerned with the problem of input-to-state stability (ISS) for a class of switched impulsive memristive neural networks with stochastic disturbance. Based on the multiple Lyapunov functions method and comparison principle, new stability results for such kind of systems are derived. The cases that all subsystems are unstable and both stable/unstable subsystems coexist are considered, respectively. Time-varying delays are taken into account in the stability analysis. The mean-square ISS delay-independent sufficient conditions are presented. Finally, numerical examples are given to illustrate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]