Search

• A complex spiking neural network (SNN) with biological rationality is constructed. • The anti-disturbance ability of SNNs with different topologies is certified. • The evolution process of neural information in SNNs under noise is clarified. The research on robustness of brain-like models contributes to promoting its neural information processing ability, and the understanding of bio-brain function. However, the biological rationality of the current brain-like models is inadequate. In addition, the effect of network topologies on the robustness of brain-like models has not been clarified. In this study, inspired by the topological characteristics of biological functional brain networks, we construct five kinds of spiking neural networks (SNNs), which have the same Izhikevich neuron model and the same synaptic plasticity model with time-delay, but different network topologies. Then, the robustness of the SNNs with different topologies is comparatively assessed based on the anti-disturbance ability under different noise. Further, by taking a speech recognition task as the case study, we investigate the anti-disturbance ability of these SNNs in application. Finally, the anti-disturbance mechanism of the SNNs is discussed. Our simulation consistently certifies that: (i) In terms of anti-disturbance indicators, the complex SNN outperforms the small-world SNN, the small-world SNN outperforms the scale-free SNN, and all of them outperform the random SNN and the regular SNN, which indicates that the SNNs with more biological rationality have the better anti-disturbance ability. (ii) In terms of speech recognition accuracy, the performance of SNNs with different topologies presents the consistent order with the anti-disturbance ability above. And the recognition accuracy of these SNNs under disturbance still remains almost the same, compared with these SNNs without disturbance. (iii) The evolution process of neural information is clarified, which hints that the synaptic plasticity is the intrinsic factor of anti-disturbance ability, and the network topology is a factor that affects the anti-disturbance ability at the level of performance. Our simulation results are conducive to the design of neuromorphic algorithms with robustness on analog or mixed-signal neuromorphic chips under complex noise environment. [ABSTRACT FROM AUTHOR]

• Completely model-free algorithm without requiring any system dynamics and initial admissible policy. • Online adaptive tuning law that synchronously update the weights of two different neural networks. • Guaranteed stability and convergence of the closed-loop system and the weights of neural networks. In this study, we present a novel algorithm, based on synchronous policy iteration, to solve the continuous-time infinite-horizon optimal control problem of input affine system dynamics. The integral reinforcement is measured as an excitation signal to estimate the solution to the Hamilton–Jacobi–Bellman equation. In addition, the proposed method is completely model-free, that is, no a priori knowledge of the system is required. Using the adaptive tuning law, the actor and critic neural networks can simultaneously approximate the optimal value function and policy. The persistence of excitation condition is required to guarantee the convergence of the two networks. Unlike in traditional policy iteration algorithms, the restriction of the initial admissible policy was eliminated using this method. The effectiveness of the proposed algorithm is verified through numerical simulations. [ABSTRACT FROM AUTHOR]

Aiming at the difference between a bicycle robot dynamics model based on prior knowledge and a real robot, the robot modeling problem is studied. A set of features is extracted from the dynamics model of a bicycle robot. The trigonometric functions in the features are replaced with a linear combination of several sigmoid functions to form a new set of features. An index describing the importance of features is designed based on the maximal information coefficient (MIC). The features are ranked using the feature importance index. The index of system identification fitness (FIT) is regarded as the likelihood function term in Bayesian information criterion (BIC). The penalty term of BIC is constructed with the number of both the features and weight parameters in the recurrent high order neural network (RHONN) model. A semi-empirical dynamics modeling method based on feature selection is proposed. An inverse optimal controller is designed to stabilize the bicycle robot. The effectiveness of the modeling method is verified through simulation and real robot experiments. [ABSTRACT FROM AUTHOR]

Brain-like intelligence is to simulate the structure and function of the biological brain as much as possible based on the latest findings in brain science. The biological brain has strong robustness under external attacks. In this study, two complex spiking neural networks (CSNNs) are constructed: a spiking neural network (SNN) with small-world topology and a SNN with scale-free topology, in which the nodes are Izhikevich neuron model, and the edges are the synaptic plasticity model including excitatory and inhibitory synapses. For the targeted attack, the anti-injury function of two CSNNs is comparatively analyzed. On this basis, the anti-injury mechanism of CSNN is explored. The experimental results show that: (1) Small-world SNN (SWSNN) has better performance than scale-free SNN (SFSNN) in the anti-injury ability under targeted attack on high-degree nodes, and they are similar in the anti-injury ability under targeted attack on intermediate and low-degree nodes; the results of the robustness of topology are consistent with the results of anti-injury function of CSNNs, which indicates that the anti-injury ability of two kinds of CSNNs is affected by topology. (2) The dynamic evolutions of neuron firing, synaptic weight, and topological characteristic in the information processing of CSNN have a linkage effect under targeted attack; the dynamic evolution of synaptic weight is significantly related to the anti-injury ability of CSNNs, which clues synaptic plasticity is the intrinsic factor of the anti-injury function of CSNNs. [ABSTRACT FROM AUTHOR]

Neural information coding is the foundation of information processing in the nervous system. The specificity of coding is the key issue of efficiently realizing complex cognition functions. In this study, a scale-free spiking neural network (SFSNN) is constructed based on synaptic plasticity and scale-free topology. The time coding properties of the SFSNN under different intensities of different stimulations including Gauss stimulation, impulse stimulation, and AC electric field stimulation are analyzed. On this basis, the specificity of time coding and the mechanism of time coding are studied emphatically. The experimental results show that: (1) Under each kind of stimulation, the inter-spike interval (ISI) is decreased, and the stronger the stimulation intensity is, the smaller the ISI is. (2) The time coding patterns have high similarity under same kind of stimulation, which is verified by the cosine similarity algorithm; the time coding patterns have high difference under different kind of stimulations, which is verified by the K-means clustering algorithm. Therefore, the time coding specificity is verified by the algorithms. (3) The relationship between the synaptic plasticity and time coding is established based on the Pearson correlation coefficient. The dynamic regulation of synaptic plasticity is significantly correlated with time coding under each kind of stimulation, which implies that the dynamic regulation of synaptic plasticity is the intrinsic factor of time coding mechanism. [ABSTRACT FROM AUTHOR]

Abstract: In this paper, the global asymptotic stability problem is dealt with for a class of neural networks (NNs) with time-varying delays. The activation functions are assumed to be neither monotonic, nor differentiable, nor bounded. By constructing an augmented Lyapunov functional which contains an integral term of neuron state vector, an improved delay-dependent stability criterion for delay NNs is established in terms of linear matrix inequalities (LMIs). It is shown that the obtained criterion can provide less conservative results than some existing ones. Numerical examples are given to demonstrate the applicability of the proposed approach. [Copyright &y& Elsevier]

Abstract: The problem of robust stability for neural networks with time-varying delays and parameter uncertainties is investigated in this paper. The parameter uncertainties are described to be of linear fractional form, which include the norm bounded uncertainties as a special case. By introducing a new Lyapunov–Krasovskii functional and considering the additional useful terms when estimating the upper bound of the derivative of Lyapunov functional, new delay-dependent stability criteria are established in term of linear matrix inequality (LMI). It is shown that the obtained criteria can provide less conservative results than some existing ones. Numerical examples are given to demonstrate the applicability of the proposed approach. [Copyright &y& Elsevier]

Abstract: An optimal fault tolerant control (FTC) scheme using output probability density functions (PDFs) is studied for the general stochastic continuous time systems. Being different from the classical FTC problems, the measured information is the stochastic distribution of the system output rather than its value. The control objective is to use the output PDFs to design control schemes that can compensate the fault and attenuate the disturbance. A multi-layer perceptron (MLP) neural network is applied to approximate the output PDFs, with which nonlinear principal component analysis (NLPCA) can be used to reduce the model order. For the established continuous-time weighting system with disturbances and uncertainties which is used to link the input and the weights, an LMI-based feasible FTC method is presented to assure that the fault can be well measured and compensated, where the H ∞ performance index for the uncertain error systems is optimized. [Copyright &y& Elsevier]

Clustering is an important research direction in data mining. However, there is no one clustering algorithm that can be applied efficiently in all situation. Clustering ensemble is the best way to solve the above-mentioned problems. It combines the results of multiple clustering algorithms, and the final result is significantly better than a single clustering algorithm. Although there is a lot of constraint information, the existing clustering ensemble algorithm does not utilize it. This paper uses constraint information in consensus function and proposes a Semi-supervised Selective Clustering Ensemble based on Chameleon (SSCEC) and Semi-supervised Selective Clustering Ensemble based on Ncut (SSCEN) to solve the above problem. SSCEC uses the chameleon algorithm as consensus function, and processes constraint information in subgraph partition and subgraph combining. SSCEN uses the Normalized cut algorithm as consensus function, and processes constraint information in the process of graph dichotomy. The experiment results show that our proposed two semi-supervised member selection clustering ensemble algorithms are better than other semi-supervised algorithms. [ABSTRACT FROM AUTHOR]

This paper is concerned with selective color feature for detecting salient regions. In contrast to most existing methods related to detection in one color space, the proposed algorithm pre-segments the input image into superpixels in both RGBY color space and Lab color space. Next, to calculate color contrast we not only consider the local feature, but also compute the difference between the pixel and the whole image. In the meanwhile, based on the center-surrounding scheme, a new computational model of color distribution features is presented to detect salient regions. Finally, 2D entropy is employed as an evaluation criterion to select and integrate the proper color features. Experimental results show that the proposed method outperforms the state-of-the-art methods on salient region detection. [ABSTRACT FROM AUTHOR]

Hashing technology can support large-scale recommendation very effectively due to its advantages of low storage cost and high recommendation efficiency. However, existing hashing-based recommendation methods often suffer from item cold-start problem. This is because they simply consider the user-item interaction and the single content information of the items, but the full interaction history is not always available and the single auxiliary information may be missing. To solve this issue, in this paper we propose a Binary Multi-modal Matrix Factorization (BMMF) method. First, we propose an efficient consensus multi-modal mapping to transform the heterogeneous multi-modal features to the unified factors by exploiting the complementarity of multiple modalities. Then, binary matrix factorization is simultaneously performed on the multi-modal features of the items and past user preferences to learn the compact binary codes of the users/items in a common Hamming space. In addition, inspired by the observation that similar instances often have similar binary codes within a short Hamming distance, we formulate a semantic structure regularization term to preserve the similarities of the items during the binary embedding process. Finally, we develop an effective Discrete Coordinate Descent (DCD) approach to tackle the formulated discrete hash optimization problem directly. Experiments on three publicly available real-world datasets demonstrate the superiority of the proposed method against the state-of-the-art methods. Our source codes and testing datasets are available at https://github.com/pcm1217/BMMF. [ABSTRACT FROM AUTHOR]

This paper presents a novel computational model to detect airports in optical remote sensing images (RSI). It works in a hierarchical architecture with a coarse layer and a fine layer. At the coarse layer, a target-oriented saliency model is built by combing the cues of contrast and line density to rapidly localize the airport candidate areas. Furthermore, at the fine layer, a learned condition random field (CRF) model is applied to each candidate area to perform the fine detection of the airport target. The CRF model is learned based on sparse features of local patches in a multi-scale structure and it also takes the contextual information of target into consideration. Therefore, its detection is more accurate and is robust to target scale variation. Comprehensive evaluations on RSI database from the Google Earth and comparisons with state-of-the-art approaches demonstrate the effectiveness of the proposed model. [ABSTRACT FROM AUTHOR]

Image-text matching is a vital yet challenging task in the field of vision and language. Unlike previous methods that usually adopt a symmetrical network to independently embed images and sentences into a joint latent space, we propose a novel Global-guided Asymmetric Attention Network (GAAN) to represent the two modalities more comprehensively. Specifically, we first design a Global Information-guided Transformer Encoder (GITE) to effectively mitigate the drawback of the lack of contextual information of the region features. Taking full advantage of the image global information, GITE is able to model the regional relations and region-global relations simultaneously, so as to obtain a more accurate visual representation. Then, we adopt a Textual Self-Attention (TSA) module to explore the word-word relations and produce the context-aware word representations. Finally, we deploy an Image-guided Textual Attention (ITA) module to explore the fine-grained correspondence between image regions and sentence words. By using context-aware visual information to guide textual representation learning, we can build asymmetric connections between vision and language to better exploit textual information. Experimental results on two benchmark datasets including MSCOCO and Flickr30k show that GAAN significantly surpasses state-of-the-art methods. [ABSTRACT FROM AUTHOR]

• We propose DBC-Forest to improve the confidence screening mechanism of gcForestcs. • DBC-Forest finds the instances with low accuracy in the high-confidence region. • DBC-Forest has better accuracy and speed performance than gcForest and gcForestcs. As a deep learning model, deep confidence screening forest (gcForestcs) has achieved great success in various applications. Compared with the traditional deep forest approach, gcForestcs effectively reduces the high time cost by passing some instances in the high-confidence region directly to the final stage. However, there is a group of instances with low accuracy in the high-confidence region, which are called mis-partitioned instances. To find these mis-partitioned instances, this paper proposes a deep binning confidence screening forest (DBC-Forest) model, which packs all instances into bins based on their confidences. In this way, more accurate instances can be passed to the final stage, and the performance is improved. Experimental results show that DBC-Forest achieves highly accurate predictions for the same hyperparameters and is faster than other similar models to achieve the same accuracy. [ABSTRACT FROM AUTHOR]

Abstract: In this paper, an anti-disturbance fault diagnosis scheme is proposed for non-Gaussian stochastic distribution systems (SDSs) with multiple disturbances. The available driven information for fault diagnosis is probability density functions (PDFs) of output rather than output value. Using B-spline expansion technique, the output PDFs can be approximated in terms of dynamic weights of B-spline neural network by which a nonlinear model can be established between input and weights. Therefore, the concerned problem is transformed into fault diagnosis problem of the weighting system presented by an uncertain nonlinear system with multiple disturbances and time-varying fault. Different from most of the existing results, the multiple disturbances are supposed to include unknown disturbance modeled by an exo-system and norm bounded uncertain disturbances. In the proposed approach, a disturbance observer is designed to estimate and compensate the modeled disturbance, and optimization technology is applied to attenuate the norm bounded disturbance. Finally, simulation results are given to show the efficiency of the proposed approach. [Copyright &y& Elsevier]

Abstract: The impulsive control method is developed to stabilize a class of Cohen–Grossberg neural networks (CGNNs) with reaction–diffusion terms and Dirichlet boundary conditions. With the help of Lyapunov functionals, impulsive delay differential inequality and comparison principle, some sufficient conditions ensuring global exponential stability (GES) of equilibrium point are obtained. In particular, the estimate of the exponential convergence rate is also provided, which depends on system parameters, time delays and impulsive interval. Numerical simulations are presented finally to demonstrate the effectiveness of proposed results. [Copyright &y& Elsevier]

Abstract: In the present paper, a new impulsive parabolic complex network (IPCN) with time-varying delays is proposed. Then, by constructing appropriate Lyapunov functionals and utilizing inequality techniques, some sufficient conditions are given to ensure that the proposed network model is globally exponentially stable. Furthermore, when the parameter uncertainties appear in the IPCN, several robust global exponential stability conditions are also presented. Finally, two examples with numerical simulations are provided to demonstrate the effectiveness of our criteria. [Copyright &y& Elsevier]

This paper studies the problem of fuzzy normalization and stabilization for a class of rectangular descriptor systems in Takagi-Sugeno (T-S) fuzzy models. It delivers a feasible scheme for the design of proportional and derivative type dynamic compensator which ensures the closed-loop system normalized and admissible. The dynamic compensator parameters are computed by solving a set of quadratic matrix inequalities, and accordingly, an efficient algorithm is built to solve related matrix inequalities. Illustrative examples are given to show the effectiveness of the present results. [ABSTRACT FROM AUTHOR]