Neural Information Processing

In this study, an adaptive support vector regressor (SVR) controller which has previously been proposed [1] is applied to control the liquid level in a spherical tank system. The variations in the cross sectional area of the tank depending on the liquid level is the main cause of nonlinearity in system. The parameters of the controller are optimized depending on the future behaviour of the system which is approximated via a seperate online SVR model of the system. In order to adjust controller parameters, the “closed-loop margin” which is calculated using the tracking error has been optimized. The performance of the proposed method has been examined by simulations carried out on a nonlinear spherical tank system, and the results reveal that the SVR controller together with SVR model leads to good tracking performance with small modeling, transient state and steady state errors.

In this paper, we develop an adaptive dynamic programming-based robust tracking control for a class of continuous-time matched uncertain nonlinear systems. By selecting a discounted value function for the nominal augmented error system, we transform the robust tracking control problem into an optimal control problem. The control matrix is not required to be invertible by using the present method. Meanwhile, we employ a single critic neural network (NN) to approximate the solution of the Hamilton-Jacobi-Bellman equation. Based on the developed critic NN, we derive optimal tracking control without using policy iteration. Moreover, we prove that all signals in the closed-loop system are uniformly ultimately bounded via Lyapunov’s direct method. Finally, we provide an example to show the effectiveness of the present approach.

Video contains a large number of motion information. The video– particularly video with moving camera – is segmented based on the relative motion occurring between moving targets and background. By using fusion ability of pulse coupled neural network (PCNN), the target regions and the background regions are fused respectively. Firstly using PCNN fuses the direction of the optical flow fusing, and extracts moving targets from video especially with moving camera. Meanwhile, using phase spectrums of topological property and color pairs (red/green, blue/yellow) generates attention information. Secondly, our video attention map is obtained by means of linear fusing the above features (direction fusion, phase spectrums and magnitude of velocity), which adds weight for each information channel. Experimental results shows that proposed method has better target tracking ability compared with three other methods– Frequency-tuned salient region detection (FT) [5], visual background extractor (Vibe) [6] and phase spectrum of quaternion Fourier transform (PQFT) [1].

We propose an exploratory form of motor babbling that uses variance predictions from a recurrent neural network as a method to acquire the body dynamics of a robot with flexible joints. In conventional research methods, it is difficult to construct real robots because of the large number of motor babbling motions required. In motor babbling, different motions may be easy or difficult to predict. The variance is large in difficult-to-predict motions, whereas the variance is small in easy-to-predict motions. We use a Stochastic Continuous Timescale Recurrent Neural Network to predict the accuracy and variance of motions. Using the proposed method, a robot can explore motions based on variance. To evaluate the proposed method, experiments were conducted in which the robot learns crank turning and door opening/closing tasks after exploring its body dynamics. The results show that the proposed method is capable of efficient motion generation for any given motion tasks.

Nonlinear time-delayed multi-agent systems with connectivity preservation are investigated in this paper. For each agent, the distributed controller is divided into five different parts which are designed to meet the requirements of the nonlinear time-delayed multi-agent systems, such as preserving connectivity, learning the unknown dynamics, eliminating time delays and reaching consensus. In addition, a $$\sigma $$-function technique is utilized to avoid the singularity in the developed distributed controller. Finally, simulation results demonstrate the effectiveness of the developed control protocol.

Rapid intensification in tropical cyclones occur where there is dramatic change in wind-intensity over a short period of time. Recurrent neural networks trained using cooperative coevolution have shown very promising performance for time series prediction problems. In this paper, they are used for prediction of rapid intensification in tropical cyclones in the South Pacific region. An analysis of the tropical cyclones and the occurrences of rapid intensification cases is assessed and then data is gathered for recurrent neural network for rapid intensification predication. The results are promising that motivate the implementation of the system in future using cloud computing infrastructure linked with mobile applications to create awareness.

This paper presents a new method of nonlinear finetwork with Gaussian kernel functions. In practice, signal enhancement filters are usually adopted as a preprocessor of signal processing system. For this purpose, an approach of nonlinear filtering using a network with Gaussian kernel functions is proposed for the efficient enhancement of noisy signals. In this method, the condition for signal enhancement is obtained by using the phase space analysis of signal time series. Then, from this analysis, the structure of nonlinear filter is determined and a network with Gaussian kernel functions is trained in such a way of obtaining the clean signal. This procedure can be repeated to obtain the multilayer (or deep) structure of nonlinear filters. As a result, the proposed nonlinear filter has demonstrated significant merits in signal enhancement compared with other conventional preprocessing filters.

In this paper, we model the skyline queries on uncertain time series, and develop a two-step procedure to answer the probabilistic skyline queries on uncertain time series. First, two effective pruning techniques are proposed to obtain the skyline in the interval. Next, two simple methods are proposed to compute the probability of each uncertain time series in the skyline. Experiments verify the effectiveness of probabilistic skylines and the efficiency and scalability of our algorithms.

This paper presents a method for probabilistic prediction of chaotic time series. So far, we have developed several model selection methods for chaotic time series prediction, but the methods cannot estimate the predictable horizon of predicted time series. Instead of using model selection methods employing the estimation of mean square prediction error (MSE), we present a method to obtain a probabilistic prediction which provides a prediction of time series and the estimation of predictable horizon. The method obtains a set of plausible predictions by means of using the similarity of attractors of training time series and the time series predicted by a number of learning machines with different parameter values, and then obtains a smaller set of more plausible predictions with longer predictable horizons estimated by LOOCV (leave-one-out cross-validation) method. The effectiveness and the properties of the present method are shown by means of analyzing the result of numerical experiments.

Time series data are generated from almost every domain and anomaly detection becomes extremely important in the last decade. It consists in detecting anomalous patterns through identifying some new and unknown behaviors that are abnormal or inconsistent relative to most of the data. An efficient anomaly detection algorithm has to adapt the detection process for each system condition and each time series behavior. In this paper, we propose an adaptive threshold able to detect anomalies in univariate time series. Our algorithm is based on segmentation and local means and standard deviations. It allows us to simplify time series visualization and to detect new abnormal data as time series jumps within different time series behavior. On synthetic and real datasets the proposed approach shows good ability in detecting abnormalities.

A major concern in cooperative coevolution for neuro- evolution is the appropriate problem decomposition method that takes into account the architectural properties of the neural network. Decomposition to the synapse and neuron level has been proposed in the past that have their own strengths and limitations depending on the application problem. In this paper, a new problem decomposition method that combines neuron and synapse level is proposed for feedfoward networks and applied to time series prediction. The results show that the proposed approach has improved the results in selected benchmark data sets when compared to related methods. It also has promising performance when compared to other computational intelligence methods from the literature.

Prediction interval (PI) is a promising tool for quantifying uncertainties associated with point predictions. Despite its informativeness, the design and deployment of PI-based controller for complex systems is very rare. As a pioneering work, this paper proposes a framework for design and implementation of PI-based controller (PIC) for nonlinear systems. Neural network (NN)-based inverse model within internal model control structure is used to develop the PIC. Firstly, a PI-based model is developed to construct PIs for the system output. This model is then used as an online estimator for PIs. The PIs from this model are fed to the NN inverse model along with other traditional inputs to generate the control signal. The performance of the proposed PIC is examined for two case studies. This includes a nonlinear batch polymerization reactor and a numerical nonlinear plant. Simulation results demonstrated that the proposed PIC tracking performance is better than the traditional NN-based controller.

We present a practical bias correction method for classifier and regression models learning under a general class of selection bias. The method hinges on two assumptions: (1) a feature vector, $${X_s}$$, exists such that S, the variable that controls the inclusion of the samples in the training set, is conditionally independent of (X, Y) given $${X_s}$$; (2) one has access to some external samples drawn from the population as a whole in order to approximate the unbiased distribution of $${X_s}$$. This general framework includes covariate shift and prior probability shift as special cases. We first show how importance weighting can remove this bias. We also discuss the case where our key assumption about $${X_s}$$ is not valid and where $${X_S}$$ is only partially observed in the test set. Experimental results on synthetic and real-world data demonstrate that our method works well in practice.

The Lagrange programming neural network (LPNN) is a framework for solving constrained nonlinear programm problems. But it can solve differentiable objective/contraint functions only. As the $$l_1$$-norm constrained quadratic minimization (L1CQM), one of the sparse approximation problems, contains the nondifferentiable constraint, the LPNN cannot be used for solving L1CQM. This paper formulates a new LPNN model, based on introducing hidden states, for solving the L1CQM problem. Besides, we discuss the stability properties of the new LPNN model. Simulation shows that the performance of the LPNN is similar to that of the conventional numerical method.

Problem decomposition is an important attribute of cooperative coevolution that depends on the nature of the problems in terms of separability which is defined by the level of interaction amongst decision variables. Recent work in cooperative coevolution featured competition and collaboration of problem decomposition methods that was implemented as islands in a method known as competitive island cooperative coevolution (CICC). In this paper, a multi-island competitive cooperative coevolution algorithm (MICCC) is proposed in which several different problem decomposition strategies are given a chance to compete, collaborate and motivate other islands while converging to a common solution. The performance of MICCC is evaluated on eight different benchmark functions and are compared with CICC where only two islands were utilized. The results from the experimental analysis show that competition and collaboration of several different island can yield solutions with a quality better than the two-island competition algorithm (CICC) on most complex multi-modal problems.

In this paper, we investigate the performance of introducing competition in cooperative coevolutionary algorithms to solve large-scale fully-separable continuous optimization problems. It may seem that solving large-scale fully-separable functions is trivial by means of problem decomposition. In principle, due to lack of variable interaction in fully-separable problems, any decomposition is viable. However, the decomposition strategy has shown to have a significant impact on the performance of cooperative coevolution on such functions. Finding an optimal decomposition strategy for solving fully-separable functions is laborious and requires extensive empirical studies. In this paper, we use a competitive two-island cooperative coevolution in which two decomposition strategies compete and collaborate to solve a fully-separable problem. Each problem decomposition has features that may be beneficial at different stages of optimization. Therefore, competition and collaboration of such decomposition strategies may eliminate the need for finding an optimal decomposition. The experimental results in this paper suggest that competition and collaboration of suboptimal decomposition strategies of a fully-separable problem can generate better solutions than the standard cooperative coevolution with standalone decomposition strategies. We also show that a decomposition strategy that implements competition against itself can also improve the overall optimization performance.

Latent Dirichlet Allocation (LDA) model is biased to draw high-frequency words to describe topics. This affects the accuracy of the representation of topics. To solve this issue, we use point-wise mutual information (PMI) to estimate the internal correlation between words and documents and propose the LDA model based on PMI. The proposed model draws words in a topic according to the mutual information. We also propose three measures to evaluate the quality of topics, which are readability, consistency of topics, and similarity of topics. The experimental results show that the quality of the topics generated by the proposed topic model is better than that of the LDA model.

Graphic Processing Unit (GPU) can achieve remarkable performance for dataset-oriented application such as Back Propagation Network (BPN) under reasonable task decomposition and memory optimization. However, advantages of GPU’s memory architecture are still not fully exploited to parallel BPN. In this paper, we develop and analyze a parallel implementation of a back propagation neural network using CUDA. It focuses on kernels optimization through the use of shared memory and suitable blocks dimensions. The implementation was tested with seven well-known benchmark data sets and the results show promising 33.8x to 64.3x speedups can be realized compared to a sequential implementation on a CPU.

In this paper, a new objective function of ICA is proposed by a probabilistic approach to the quadratic terms. Many previous ICA methods are sensitive to the sign of kurtosis of source (sub- or super-Gaussian), where the change of the sign often causes a large discontinuity in the objective function. On the other hand, some other previous methods use continuous objective functions by using the squares of the 4th-order statistics. However, such squared statistics often lack the robustness because they magnify the outliers. In this paper, we solve this problem by introducing a new objective function which is given as a summation of weighted 4th-order statistics, where the kurtoses of sources are incorporated “smoothly” into the weights. Consequently, the function is always continuously differentiable with respect to both the kurtoses and the separating matrix to be estimated. In addition, we propose a new ICA method optimizing the objective function by the Givens rotations under the orthonormality constraint. Experimental results show that the proposed method is comparable to the other ICA methods and it outperforms them especially when sub-Gaussian sources are dominant.

A cascaded factor analysis network is proposed in this paper, which is suitable for extracting distributed semantic representations to various problems ranging from digit recognition and image classification to face recognition. There are two key points in this novel model: 1. simplify and accelerate the deep convolution networks with competitive accuracy even state-of-the-art for many general image tasks; 2. combine a statistical methodfactor analysis with neural networks for excellent automatically learning ability and abundant semantic information. Experiments on many benchmark visual datasets demonstrate that this simple network performs efficiently and effectively while attaining competitive accuracy to the current state-of-the-art methods.

Open Vehicle routing problems is a variant of Vehicle Routing Problem, in which vehicles don’t return the depot after serving the customers. In this study, we proposed a cluster first-routed second based algorithm. We combined Kmeans and Variable Neighborhood Search in this algorithm. Our proposed algorithm achieves the best know solutions within a reasonable time for all well-known small and medium scale benchmarks.

A common measurement model for locating a mobile source is time-of-arrival (TOA). However, when non-line-of-sight (NLOS) bias error exists, the error can seriously degrade the estimation accuracy. This paper formulates the problem of estimating a mobile source position under the NLOS situation as a nonlinear constrained optimization problem. Afterwards, we apply the concept of Lagrange programming neural networks (LPNNs) to solve the problem. In order to improve the stability at the equilibrium point, we add an augmented term into the LPNN objective function. Simulation results show that the proposed method provides much robust estimation performance.

We propose wave-based computing based on coupled oscillators to avoid the inter-connection bottleneck in large scale and densely integrated cognitive systems. In addition, we introduce the concept of reservoir computing to coupled oscillator systems for non-conventional physical implementation and reduction of the training cost of large and dense cognitive systems. We show that functional approximation and regression can be efficiently performed by synchronization of coupled oscillators and subsequent simple readouts.

This work presents a hybrid controller based on the combination of fuzzy logic control (FLC) mechanism and internal model-based control (IMC). Neural network-based inverse and forward models are developed for IMC. After designing the FLC and IMC independently, they are combined in parallel to produce a single control signal. Mean averaging mechanism is used to combine the prediction of both controllers. Finally, performance of the proposed hybrid controller is studied for a nonlinear numerical plant model (NNPM). Simulation result shows the proposed hybrid controller outperforms both FLC and IMC.

With the flood of publicly available data, it allows scientists to explore and discover new findings. Gene expression is one type of biological data which captures the activity inside the cell. Studying gene expression data may expose the mechanisms of disease development. However, with the limitation of computing resources or knowledge in computer programming, many research groups are unable to effectively utilize the data. For about a decade now, various web-based data analysis tools have been developed to analyze gene expression data. Different tools were implemented by different analytical approaches, often resulting in different outcomes. This study conducts a comparative study of three existing web-based gene expression analysis tools, namely Gene-set Activity Toolbox (GAT), NetworkAnalyst and GEO2R using six publicly available cancer data sets. Results of our case study show that NetworkAnalyst has the best performance followed by GAT and GEO2R, respectively.

It is often said that the eyes are the windows to the soul. If that is true, then it may also be inferred that looking at web users’ eye movements could potentially reflect what they are actually thinking when they view websites. In this paper, we conduct a set of experiments to analyze whether user intention in relation to assessing the credibility of a website can be extracted from eye movements. In our within-subject experiments, the participants determined whether twenty websites seemed to be phishing websites or not. We captured their eye movements and tried to extract intention from the number and duration of eye fixations. Our results demonstrated the possibility to estimate a web user’s intention when making a trust decision, solely based on the user’s eye movement analysis.

We present new image features for diagnosing general wellbeing states and medical conditions. The new method, called Gaussian Hamming Distance (GHD), generates de-identified features that are highly correlated with general wellbeing states, such as happiness, smoking, and facial palsy. This method allows aid organizations and governments in developing countries to provide affordable medical services. We evaluate the new approach using real face-image data and four classifiers: Naive Bayesian classier, Artificial Neural Network, Decision Tree, and Support Vector Machines (SVM) for predicting general wellbeing states. Its predictive power (over 93 % accuracy) is suitable for providing a variety of online services including recommending useful health information for improving general wellbeing states.

This paper presents a multi-shot person re-identification system from video sequences based on Interest Points (SURFs) matching. Our objective is to improve the Interest Points (IPs) matching using low resolution images in terms of re-identification accuracy and running time. First, we propose a new method of SURF matching via Local Sparse Representation (LSR). Each SURF in the test video sequence is expressed as a sparse representation of a subset of SURFs in the reference dataset. Our approach consists of searching the latter subset from the reference IPs that are located on a similar spatial neighborhood to the query IP. Second, it investigates whether IPs filtering can decrease the re-identification running time. An ensemble of binary classifiers are evaluated. Our approach is assessed on the large dataset PRID-2011 and shown to outperform favorably with current state of the art.

This paper proposes a host (corresponding to a source IP) behavior based traffic decomposition approach to identify groups of malicious events from massive historical darknet traffic. In our approach, we segmented and extracted traffic flows from captured darknet data, and categorized flows according to a set of rules that summarized from host behavior observations. Finally, significant events are appraised by three criteria: (a) the activities within each group should be highly alike; (b) the activities should have enough significance in terms of scan scale; and (c) the group should be large enough. We applied the approach on a selection of twelve months darknet traffic data for malicious events detection, and the performance of the proposed method has been evaluated.

Artificial neural network (ANN) models are able to predict future events based on current data. The usefulness of an ANN lies in the capacity of the model to learn and adjust the weights following previous errors during training. In this study, we carefully analyse the existing methods in neuronal spike sorting algorithms. The current methods use clustering as a basis to establish the ground truths, which requires tedious procedures pertaining to feature selection and evaluation of the selected features. Even so, the accuracy of clusters is still questionable. Here, we develop an ANN model to specially address the present drawbacks and major challenges in neuronal spike sorting. New enhancements are introduced into the conventional backpropagation ANN for determining the network weights, input nodes, target node, and error calculation. Coiflet modelling of noise is employed to enhance the spike shape features and overshadow noise. The ANN is used in conjunction with a special spiking event detection technique to prioritize the targets. The proposed enhancements are able to bolster the training concept, and on the whole, contributing to sorting neuronal spikes with close approximations.

This paper presents a novel sufficient condition for the existence, uniqueness and global robust asymptotic stability of the equilibrium point for the class of delayed neural networks by using the Homomorphic mapping and the Lyapunov stability theorems. An important feature of the obtained result is its low computational complexity as the reported result can be verified by checking some well-known properties of some certain classes of matrices, which simplify the verification of the derived result.

The purpose of this paper is to discuss the use of $$L_{2}E$$ estimation that minimizes integrated square distance as a practical robust estimation tool for unsupervised clustering. Comparisons to the expectation maximization (EM) algorithm are made. The $$L_{2}E$$ approach for mixture models is particularly useful in the study of big data sets and especially those with a consistent numbers of outliers. The focus is on the comparison of $$L_{2}E$$ and EM for parameter estimation of Gaussian Mixture Models. Simulation examples show that the $$L_{2}E$$ approach is more robust than EM when there is noise in the data (particularly outliers) and for the case when the underlying probability density function of the data does not match a mixture of Gaussians.

This paper presents a novel model predictive control (MPC) approach to tracking control of mobile robots based on recurrent neural networks (RNNs). The tracking control problem is firstly formulated as a sequential dynamic optimization problem in framework of MPC. Then a novel neurodynamic approach is developed for computing the optimal control signals in real time, where multiple RNNs are applied in a collective fashion. The proposed approach enables MPC of mobile robots to be synthesized in real time. Simulation results are provided to substantiate the effectiveness of the proposed approach.

This study is concerned with the delay-range-dependent stability analysis for neural networks with time-varying delay and Markovian jumping parameters. The time-varying delay is assumed to lie in an interval of lower and upper bounds. The Markovian jumping parameters are introduced in delayed neural networks, which are modeled in a continuous-time along with finite-state Markov chain. Moreover, the sufficient condition is derived in terms of linear matrix inequalities based on appropriate Lyapunov-Krasovskii functionals and stochastic stability theory, which guarantees the globally asymptotic stable condition in the mean square. Finally, a numerical example is provided to validate the effectiveness of the proposed conditions.

The involvement of astrocytes in information processing in the brain has recently been demonstrated. In this paper, we investigate, using computational models (SOM and MLP), one of the observed astrocyte-neuron interactions for information processing: the neural modulation represented by the observed calcium waves. We apply it to solve classification problems. The results of the performed tests confirmed that the proposed approach improved artificial neural network performance, especially learning time acceleration.

This paper presents the growing greedy search algorithm and its application to associative memories of hysteresis neural networks in which storage of desired memories are guaranteed. In the algorithm, individuals correspond to cross-connection parameters, the cost function evaluates the number of spurious memories, and the set of individuals can grow depending on the global best. Performing basic numerical experiments, the algorithm efficiency is investigated.

Galaxy groups play a significant role in explaining the evolution of the universe. Given the amounts of available survey data, automated discovery of galaxy groups is of utmost interest. We introduce a novel methodology, based on probabilistic Hough transform, for finding galaxy groups embedded in a rich background. The model takes advantage of a typical signature pattern of galaxy groups known as “fingers-of-God”. It also allows us to include prior astrophysical knowledge as an inherent part of the method. The proposed method is first tested in large scale controlled experiments with 2-D patterns and then verified on 3-D realistic mock data (comparing with the well-known friends-of-friends method used in astrophysics). The experiments suggest that our methodology is a promising new candidate for galaxy group finders developed within a machine learning framework.

Evolutionary algorithms have been frequently applied to constrained continuous optimisation problems. We carry out feature based comparisons of different types of evolutionary algorithms such as evolution strategies, differential evolution and particle swarm optimisation for constrained continuous optimisation. In our study, we examine how sets of constraints influence the difficulty of obtaining close to optimal solutions. Using a multi-objective approach, we evolve constrained continuous problems having a set of linear and/or quadratic constraints where the different evolutionary approaches show a significant difference in performance. Afterwards, we discuss the features of the constraints that exhibit a difference in performance of the different evolutionary approaches under consideration.

Different types of evolutionary algorithms have been developed for constrained continuous optimisation. We carry out a feature-based analysis of evolved constrained continuous optimisation instances to understand the characteristics of constraints that make problems hard for evolutionary algorithm. In our study, we examine how various sets of constraints can influence the behaviour of $$\varepsilon $$-Constrained Differential Evolution. Investigating the evolved instances, we obtain knowledge of what type of constraints and their features make a problem difficult for the examined algorithm.

In this research paper, a novel Convolutional Associative Memory is proposed. In the proposed model, Synapse of each neuron is modeled as a Linear FIR filter. The dynamics of Convolutional Associative Memory is discussed. A new method called Sub-sampling is given. Proof of convergence theorem is discussed. An example depicting the convergence is shown. Some potential applications of the proposed model are also proposed.

As one of the most important techniques in recommender systems, collaborative filtering (CF) generates the recommendations or predictions based on the observed preferences. Most traditional recommender systems fail to discover the latent associations between the same or similar items with different names, which is called synonymy problem. With the rapid increasing number of users and items, the user-item rating data is extremely sparse. Based on the limited number of user ratings, we cannot capture enough information from the user history using the traditional CF techniques, which could reduce the effectiveness of the recommender systems.In this paper, we propose a novel model User-Relation-Item Model (URIM) for CF, which exploits the latent relationship between different user interest domains and item types. By introducing a component named user-item-relation matrix, which reflects the latent major association patterns behind users and items, URIM tackles the synonymy problem, and therefore achieves a significant performance improvement. We compared our method with several state-of-the-art recommendation algorithms on two real-world datasets. Experimental results validate the effectiveness of our model in terms of prediction accuracy (RMSE) and top-N recommendation quality (Recall and Precision). More specifically, URIM reduces the RMSE by nearly 10 % and 5 % on the two datasets, respectively.

Collaborative filtering (CF) systems aim at recommending a set of personalized items for an active user, according to the preferences of other similar users. Many methods have been developed and some, such those based on Similarity and Matrix Factorization (MF) can achieve very good recommendation accuracy, but unfortunately they are computationally prohibitive. Thus, applying such approaches to real-world applications in which available information evolves frequently, is a non-trivial task. To address this problem, we propose a novel efficient incremental CF system, based on a weighted clustering approach. Our system is able to provide a high quality of recommendations with a very low computation cost. Experimental results on several real-world datasets, confirm the efficiency and the effectiveness of our method by demonstrating that it is significantly better than existing incremental CF methods in terms of both scalability and recommendation quality.

Static and dynamic taint-analysis approaches have been developed to detect the processing of sensitive information. Unfortunately, faced with the result of analysis about operations of sensitive information, people have no idea of which operation is legitimate operation and which is stealthy malicious behavior. In this paper, we present Monkeydroid to pinpoint automatically whether the android application would leak sensitive information of users by distinguishing the reasonable and unreasonable operation of sensitive information on the basis of information provided by developer and market provider. We evaluated Monkeydroid over the top 500 apps on the Google play and experiments show that our tool can effectively distinguish malicious operations of sensitive information from legitimate ones.

This paper presents a new approach to segment and track people in video. The basic idea is the use of deformable model with incorporation of statistical prior. We propose an hybrid energy model that incorporates a global and a statistical based energy terms in order to improve the tracking task even under occlusion conditions. Target models are initialized at the first frame, then predictions are constructed based on motion vectors. Therefore, we apply an hybrid active contour model in order to segment tracked people. Experiments show the ability of the proposed algorithm to detect, segment and track people well.

We propose in this paper a new approach to detect and visualize the change in a streaming clustering. This approach can be used to explore visually the data streams. We assume that the data stream structure can be different during the time. Our objective is to alert the user on the structure change during the time period. A common approach to deal with data streams is to observe and process it in a window. The principle of the proposed approach is to apply a data exploration method on each window. We then propose to visualize the change between all windows for each extracted cluster. The user can investigate more precisely the change between the two windows through a visual projection for each extracted cluster.

Salient objects detection aims to locate objects that capture human attention within images. Recent progresses in saliency detection have exploited the center prior, to combine with other cues such as background information, object size or region contrast, achieving competitive results. However, previous approaches of center prior supposing salient object locates nearly at image center is very simple, fragile, especially not suitable for multiple salient objects detection, but the assumption is mostly heuristic. In this paper, we present an adaptive location method based on geodesic filtering framework to address these issues. First, we detect salient points by the adjustive color Harris algorithm. Second, we involve the Affinity Propagation (AP) method to automatically cluster the salient points for a coarse objects location. Then, we utilize geodesic filtering framework for a final saliency map by multiplying objects location and size. Experimental results on two more challenging databases of off-center and multiple salient objects demonstrate our approach is more robust to the location variations of salient objects, against state-of-the-art methods for saliency detection.

The problem of anomaly detection is a critical topic across application domains and is the subject of extensive research. Applications include finding frauds and intrusions, warning on robot safety, and many others. Standard approaches in this field exploit simple or complex system models, created by experts using detailed domain knowledge.In this paper, we put forth a statistics-based anomaly detector motivated by the fact that anomalies are sparse by their very nature. Powerful sparsity directed algorithms—namely Robust Principal Component Analysis and the Group Fused LASSO—form the basis of the methodology. Our novel unsupervised single-step solution imposes a convex optimisation task on the vector time series data of the monitored system by employing group-structured, switching and robust regularisation techniques.We evaluated our method on data generated by using a Baxter robot arm that was disturbed randomly by a human operator. Our procedure was able to outperform two baseline schemes in terms of $$F_1$$ score. Generalisations to more complex dynamical scenarios are desired.

In this paper, we propose an Appearance and Shape Constrained Active Basis Model (ASC-ABM) to detect vehicles in image. ASC-ABM effectively incorporates the appearance and shape prior of vehicles in the active basis model. Therefore, compared with the original ABM, it can effectively remove the false positives caused by the clutter background and traffic lines. Experiment results demonstrate the effectiveness of the proposed method.

Clustering or cluster analysis is an important and common task in data mining and analysis, with applications in many fields. However, most existing clustering methods are sensitive in the presence of limited amounts of data per cluster in real-world applications. Here we propose a new method called denoising cluster analysis to improve the accuracy. We first construct base clusterings with artificially corrupted data samples and later learn their ensemble based on mutual information. We develop multiplicative updates for learning the aggregated cluster assignment probabilities. Experiments on real-world data sets show that our method unequivocally improves cluster purity over several other clustering approaches.

Image de-noising is one of the important information processing technologies and a fundamental image processing step for improving the overall quality of medical images. Conventional de-noising methods, however, tend to over-suppress high-frequency details. To overcome this problem, in this paper we present a novel compressive sensing (CS) based noise removing algorithm using proposed sparse basis on CDF9/7 wavelet transform. The measurement matrix is applied to the transform coefficients of the noisy image for compressive sampling. The orthogonal matching pursuit (OMP) and Basis Pursuit (BP) are applied to reconstruct image from noisy sparse image. In the reconstruction process, the proposed threshold with Bayeshrink thresholding strategies is used. Experimental results demonstrate that the proposed method removes noise much better than existing state-of-the-art methods in the sense image quality evaluation indexes.

Trajectory abstracting is to compendiously summarize the substance of a lot of information delivered by the trajectory data. In this paper, to cope with complex trajectory data, we propose a novel framework for abstracting trajectories from the perspective of signal processing. That is, trajectories are designated as signals, manifesting the copious information that varies with time and space, and denoising is exploited to concisely communicate the trajectory data. Resampling of trajectory data is firstly performed, based on achieving the minimum Jensen-Shannon divergence of the trajectories before and after being re-sampled. The resampled trajectories are matched into groups according to their similarity and, a non-local denoising approach based on wavelet transformation is developed to produce summaries of trajectory groups. Our new framework can not only offer multi-granularity abstractions of trajectory data, but also identify outlier trajectories. Extensive experimental studies have shown that the proposed framework achieves very potential results in trajectory summarization, in terms of both objective evaluation metrics and subjective visual effects. To the best of our knowledge, this is the first to deploy the group-based signal denoising technique in the context of summarizing the trajectory data.

Sparse representation models code image patches as a linear combination of a few atoms selected from a given dictionary. Sparse representation-based image denoising (SRID) models, learning an adaptive dictionary directly from the noisy image itself, has shown promising results for image denoising. However, due to the noise of the observed image, these conventional models cannot obtain good estimations of sparse coefficients and the dictionary. To improve the performance of SRID models, we propose a multi-scale fractional-order sparse representation (MFSR) model for image denoising. Firstly, a novel sample space is re-estimated by respectively correcting singular values with the non-linear fractional-order technique in wavelet domain. Then, the denoised image can be reconstructed with the accurate sparse coefficients and optimal dictionary in the novel sample space. Compared with the conventional SRID models and other state-of-the-art image denoising algorithms, the experimental results show that the performances of our proposed MFSR model are much better in terms of the accuracy, efficiency and robustness.

A novel PDE-based image restoration approach is proposed in this article. The provided PDE model is based on a linear second-order hyperbolic diffusion equation. The well-posedness of the proposed differential model and some nonlinear PDE schemes derived from it are also discussed. A consistent and fast-converging numerical approximation scheme using finite differences is then constructed for the continuous hyperbolic PDE model. Some image restoration experiments using this approach and several method comparisons are also described.

In this paper, we study the effect of noise on a gradient system with forgetting. The noise include multiplicative noise, additive noise and chaotic noise. For multiplicative or additive noise, the noise is a mean zero Gaussian noise. It is added to the state vector of the system. For chaotic noise, it is added to the gradient vector. Let $${\mathbf x}$$ be the state vector of a system, $$S_b$$ be the variance of the Gaussian noise, $$\kappa '$$ is average noise level of the chaotic noise, $$\lambda $$ is a positive constant, $$V({\mathbf x})$$ be the energy function of the original gradient system, $$V_{\otimes }({\mathbf x})$$, $$V_{\oplus }({\mathbf x})$$ and $$V_{\odot }({\mathbf x})$$ be the energy functions of the gradient systems, if multiplicative, additive and chaotic noises are introduced. Suppose $$V({\mathbf x}) = F({\mathbf x}) + \lambda \Vert {\mathbf x}\Vert ^2_2$$. It is shown that $$V_{\otimes }({\mathbf x}) = V({\mathbf x}) + (S_b/2) \sum _{j=1}^n (\partial ^2 F({\mathbf x})/\partial x_j^2) x_j^2 - S_b \sum _{j=1}^n \int x_j (\partial ^2 F({\mathbf x})/\partial x_j^2) dx_j$$, $$V_{\oplus }({\mathbf x}) = V({\mathbf x}) + (S_b/2) \sum _{j=1}^n \partial ^2 F({\mathbf x})/\partial x_j^2$$, and $$V_{\odot }({\mathbf x}) = V({\mathbf x}) + \kappa '\sum _{i=1}^n x_i$$. The first two results imply that multiplicative or additive noise has no effect on the system if $$F({\mathbf x})$$ is quadratic. While the third result implies that adding chaotic noise can have no effect on the system if $$\kappa '$$ is zero. As many learning algorithms are developed based on the method of gradient descent, these results can be applied in analyzing the effect of noise on those algorithms.

Advances in Web 2.0 technology has led to the popularity of social networking sites. One fundamental task for social networking sites is to recommend appropriate new friends for users. In recent years, network structure has been used for user recommendation. Most existing network structure-based recommendation methods either need to pre-specify the group number and structure type or fail to improve performance. In this paper, we propose a novel network structure-based user recommendation method, called Bayesian nonparametric mixture matrix factorization (BNPM-MF). The BNPM-MF model first employs a Bayesian nonparametric model to automatically determine the group number and the network structure in networks and then applies a matrix factorization method on each structure to user recommendation for improvement. Experiments conducted on a number of real networks demonstrate that the BNPM-MF model is competitive with other state-of-the-art methods.

A set of genes and their regulatory interactions are represented in a gene regulatory network (GRN). Since GRNs play a major role in maintaining the cellular activities, inferring these networks is significant for understanding biological processes. Among the models available for GRN reconstruction, our recently developed nonlinear model [1] using Michaelis-Menten kinetics is considered to be more biologically relevant. However, the model remains coupled in the current form making the process computationally expensive, especially for large GRNs. In this paper, we enhance the existing model leading to a decoupled form which not only speeds up the computation, but also makes the model more realistic by representing the strength of each regulatory arc by a distinct Michaelis-Menten constant. The parameter estimation is carried out using differential evolution algorithm. The model is validated by inferring two synthetic networks. Results show that while the accuracy of reconstruction is similar to the coupled model, they are achieved at a faster speed.

Overfitting is an important problem in neural networks (NNs) training. When the number of samples in the training set is limited, explicitly extending the training set with artificially generated samples is an effective solution. However, this method has the problem of high computational costs. In this paper we propose a new learning scheme to train single-hidden layer feedforward neural networks (SLFNs) with implicitly extended training set. The training set is extended by corrupting the hidden layer outputs of training samples with noise from exponential family distribution. When the number of corruption approaches infinity, in objective function explicitly generated samples can be expressed as the form of expectation. Our method, called marginalized corrupted hidden layer (MCHL), trains SLFNs by minimizing the loss function expected values under the corrupting distribution. In this way MCHL is trained with infinite samples. Experimental results on multiple data sets show that MCHL can be trained efficiently, and generalizes better to test data.

Ensemble learning algorithms aim to train a group of classifiers to enhance the generalization ability. However, vast of those algorithms are learning in batches and the base classifiers (e.g. number, type) must be predetermined. In this paper, we propose an ensemble algorithm called INLEX (Incremental Network with Local EXperts ensemble) to learn suitable number of linear classifiers in an online incremental mode. Specifically, it incrementally learns the representational nodes of the input space. In the incremental process, INLEX finds nodes in the decision boundary area (boundary nodes) based on the theory of entropy: boundary nodes are considered to be disordered. In this paper, boundary nodes are activated as experts, each of which is a local linear classifier. Combination of these linear experts with dynamical weights will constitute a decision boundary to solve nonlinear classification tasks. Experimental results show that INLEX obtains promising performance on real-world classification benchmarks.

The volume transmission (VT), a new type of cellular signaling, is based on the diffusion of neuro-active substances such as Nitric Oxide (NO) in the Extracellular Space (ECS). It is not homogeneous, critically dependent on, and limited by, its structure and physico-chemical properties. We present a different computational model of the NO diffusion based on multi-compartmental systems and transportation phenomena. It allows incorporating these ECS characteristics and the biological features and restrictions of the NO dynamics.This discrete model will allow to determine the NO dynamics and its capabilities in cellular communication and formation of complex structures in biological and artificial environments.This paper addresses the design model and its analysis in one-dimensional and three-dimensional environment, over trapezoidal generation and diffusion processes.

Multidimensional networks, networks with multiple kinds of relations, widely exist in various fields. Structure exploration (i.e., structural regularity exploration) is one fundamental task of network analysis. Most existing structural regularity exploration methods for multidimensional networks need to pre-assume which type of structure they have, and some methods that do not need to pre-assume the structure type usually perform poorly. To explore structural regularities in multidimensional networks well without pre-assuming which type of structure they have, we propose a novel feature aggregation method based on a mixture model and Bayesian theory, called the multidimensional Bayesian mixture (MBM) model. Experiments conducted on a number of synthetic and real multidimensional networks show that the MBM model achieves better performance than other relative models on most networks.

Multipath fading is one of the most serious problems in mobile communications. Various methods to solve or mitigate it have been proposed in time or frequency domain. Previously we proposed a channel prediction method that combines complex-valued neural networks and chirp z-transform that utilizes both the time- and frequency-domain representation, resulting in much higher performance. In this paper, we propose to deal with polarization additionally in its adaptive channel prediction to improve the performance further. A preliminary experiment demonstrates improvement larger than what is expected by a simple diversity gain.

The low rank matrix completion problem has attracted great attention and been widely studied in collaborative filtering and recommendation systems. The rank minimization problem is NP-hard, so the problem is usually relaxed into a matrix nuclear norm minimization. However, the usage is limited in scability due to the high computational complexity of singular value decomposition (SVD). In this paper we introduce a random projection to handle this limitation. In particular, we use a randomized SVD to accelerate the classical Soft-Impute algorithm for the matrix completion problem. The empirical results show that our approach is more efficient while achieving almost same performance.

In this paper, we present a Custom Parallel Architecture for Neural networks (CuPAN). CuPAN consists of streamlined nodes that each node is able to integrate a single or a group of neurons. It relies on a high-throughput and low-cost Clos on-chip interconnection network in order to efficiently handle inter-neuron communication. We show that the similarity between the traffic pattern of neural networks (multicast-based multi-stage traffic) and topological characteristics of multi-stage interconnection networks (MINs) makes neural networks naturally suited to the MINs. The Clos network, as one of the most important classes of MINs, provide scalable low-cost interconnection fabric composed of several stages of switches to connect two groups of nodes and interestingly, can support multicast in an efficient manner. Our evaluation results show that CuPAN can manage the multicast-based traffic of neural networks better than the mesh-based topologies used in many parallel neural network implementations and gives lower average message latency, which directly translates to faster neural processing.

Forecasting bike sharing demand is of paramount importance for management of fleet in city level. Rapidly changing demand in this service is due to a number of factors including workday, weekend, holiday and weather condition. These nonlinear dependencies make the prediction a difficult task. This work shows that type-1 and type-2 fuzzy inference-based prediction mechanisms can capture this highly variable trend with good accuracy. Wang-Mendel rule generation method is utilized to generate rulebase and then only current information like date related information and weather condition is used to forecast bike share demand at any given point in future. Simulation results reveal that fuzzy inference predictors can potentially outperform traditional feedforward neural network in terms of prediction accuracy.

Visual functions of the elderly are gradually changing with age. As one of the changes, aged eyes have a different property in perceiving high-frequency components from younger eyes. In general, the elderly perceives images differently from the younger. To give the same perception for the same image, a different image from an original image needs to be displayed to the elderly. In this paper, a method of generating the inverse characteristic of the image filter for the presbyopia is proposed. To this end, a serially-cascaded neural network model is proposed. The neural network is composed of 4 layers. The 4-layer neural network is divided into 2 blocks on the aspect of function. The upper and the lower layers play a role of simulating the image conversion and obtaining its opposite characteristic, respectively. The performance of the proposed framework is evaluated by the experiments on the image pre-conversion for the presbyopia.

Identification of functional connectivity between neurons is an important issue in computational neuroscience. Recently, the number of simultaneously recorded neurons is increasing, and computational complexity to estimate functional connectivity is exploding. In this study, we propose a two-stage algorithm to estimate spike response functions between neurons in a large scale network. We applied the proposed algorithm to various scales of neural networks and showed that the computational complexity is reduced without sacrificing estimation accuracy.

Recommender systems has been wildly used in many websites. These perform much better on users for which they have more information. Satisfying the needs of users new to a system has become an important problem. It is even more accurate considering that some of these hard to describe new users try out the system which unfamiliar to them by their ability to immediately provide them with satisfying recommendations, and may quickly abandon the system when disappointed. Quickly determining user preferences often through a boot process to achieve, it guides users to provide their opinions on certain carefully chosen items or categories. In particular, we advocate a matrix completion solution as the most appropriate tool for this task. We focus on online and offline algorithms that use data compression algorithm and the decision tree which has been built to do real-time recommendation. We merge the three algorithms : distributed matrix completion, cluster-based and decision-tree-based, We chose different algorithms based on different scenarios. The experimental study delivered encouraging results, with the matrix completion bootstrapping process significantly outperforming previous approaches. abstract environment.

Aiming at various tracking environments, a weighted joint sparse representation based tracker is proposed. Specifically, each object template is weighted according to its similarity to each candidate. Then all candidates are represented sparsely and jointly, and the sparse coefficients are used to compute the observation probabilities of candidates. The candidate with the maximum observation probability is determined as the object. The object function is solved by a modified accelerated proximal gradient (APG) algorithm. Experiments on several representative image sequences show that the proposed tracking method performs better than the other trackers in the scenarios of illumination variation, occlusion, pose change and rotation.

It is well known that super-resolution (SR) is a difficult problem, especially the single-frame super-resolution (SFSR). In this paper, we propose a novel SFSR method, called compressive sampling on hybrid reconstructions (CSHR), with high reconstruction quality and relatively low computation cost. It mainly depends on the combination of the results of other SR methods, which are characteristic of high speed and low quality SR results alone. As a result, CSHR inherits the merit of low computation cost. We resample those low quality SR results in DCT domain instead of in pixel domain and regard the similar expansion coefficients as consensus which would be compressively sampled later. In CSHR, obtaining a high resolution image is only to solve a convex optimization program. We use compressed sensing theory to ensure the efficiency of our method. Also, we give some theoretic results. Experimental results show the effectiveness of the proposed method when compared to some state-of-the-art methods.

More convincing evidence has proven the existence of a bidirectional relationship between neurons and astrocytes. Astrocytes, a new type of glial cells previously considered as passive support cells, constitute a system of non-synaptic transmission playing a major role in modulating the activity of neurons. In this context, this paper proposes to model the effect of these cells to develop a new type of artificial neural network operating on new mechanisms to improve the information processing and reduce learning time, very expensive in traditional networks. The obtained results indicate that the implementation of bio-inspired functions such as of astrocytes, improve very considerably learning speed.The developed model achieves learning up to twelve times faster than traditional artificial neural networks.

Changes in the skill requirements of occupations can alter the balance in the numbers of high, middle and low-skilled jobs on the market. This can result in structural unemployment, stagnating income and other unforeseen social and economic side effects. In this paper, we demonstrate the use of a recent matrix factorization technique for extracting the underlying skill categories from O*NET, a publicly available database on occupational skill requirements. This study builds upon earlier work which also focused on this database, and which indicated that changes in skill requirements were in response to increased automation which unevenly affected different segments of the job market. In this paper we refine the methodological underpinnings of the earlier work and report some preliminary results which already show great promise.

Non-negative Matrix Factorization (NMF) is proven to be a very effective decomposition method for dimensionality reduction in data analysis, and has been widely applied in computer vision, pattern recognition and information retrieval. However, NMF is virtually an unsupervised method since it is unable to utilize prior knowledge about data. In this paper, we present Constrained Non-negative Matrix Factorization with Graph Laplacian (CNMF-GL), which not only employs the geometrical information, but also properly uses the label information to enhance NMF. Specifically, we expect that a graph regularized term could preserve the local structure of original data, meanwhile data points both having the same label and possessing different labels will have corresponding constraint conditions. As a result, the learned representations will have more discriminating power. The experimental results on image clustering manifest the effectiveness of our algorithm.

Winner(s) determination in online reverse auctions is a very appealing e-commerce application. This is a combinatorial optimization problem where the goal is to find an optimal solution meeting a set of requirements and minimizing a given procurement cost. This problem is hard to tackle especially when multiple attributes of instances of items are considered together with additional constraints, such as seller’s stocks and discount rate. The challenge here is to determine the optimal solution in a reasonable computation time. Solving this problem with a systematic method will guarantee the optimality of the returned solution but comes with an exponential time cost. On the other hand, approximation techniques such as evolutionary algorithms are faster but trade the quality of the solution returned for the running time. In this paper, we conduct a comparative study of several exact and evolutionary techniques that have been proposed to solve various instances of the combinatorial reverse auction problem. In particular, we show that a recent method based on genetic algorithms outperforms some other methods in terms of time efficiency while returning a near to optimal solution in most of the cases.

Implementations of visual simulations of shock phenomenon have been given significantly less-attention in last decades. We present a novel approach to simulate aero-optical distortions due to shock waves generated by a supersonic jet by considering the physics background of the shock phenomenon. The optical distortion is simulated by calculating the index of refraction for oblique shock waves. The refractive index for the shock wave was calculated, by considering the mean characteristics of supersonic flows. Even though the flow characteristics are not uniform across the shock wave the results shows that this approach is a better way to simulate aero-optical distortions in real time.

The number of security incidents faced by Android users is growing, along with a surge in malware targeting Android terminals. Such malware arrives at the Android terminals in the form of Android Packages (APKs). Various techniques for protecting Android users from such malware have been reported, but most of them have focused on the APK files themselves. Unlike these approaches, we use Web information obtained from online APK markets to improve the accuracy of malware detection. In this paper, we propose category/cluster-based APK analysis schemes that quantify the risk of an APK. The category-based scheme uses category information available on the Web, whereas the cluster-based method uses APK descriptions to generate clusters of APK files. In this paper, the performance of the proposed schemes is verified by comparing their area under the curve values with that of a conventional scheme; moreover, the usability of Web information for the purpose of better quantifying the risks of APK files is confirmed.

Opinion formation modelling is still poorly understood due to the hardness and complexity of the abstraction of human behaviours under the presence of various types of social influences. Two such influences that shape the opinion formation process are: (i) the expert effect originated from the presence of experts in a social group and (ii) the majority effect caused by the presence of a large group of people sharing similar opinions. In real life when these two effects contradict each other, they force public opinions towards their respective directions. Existing models employed the concept of confidence levels associated with the opinions to model the expert effect. However, they ignored the majority effect explicitly, and thereby failed to capture the combined impact of these two influences on opinion evolution. Our model explicitly introduces the majority effect through the use of a concept called opinion consistency, and captures the opinion dynamics under the combined influence of majority supported opinions as well as experts’ opinions. Simulation results show that our model properly captures the consensus, polarization and fragmentation properties of public opinion and reveals the impact of the aforementioned effects.

Distributed systems are widely used for solving large-scale and data-intensive computing problems, including all-to-all comparison (ATAC) problems. However, when used for ATAC problems, existing computational frameworks such as Hadoop focus on load balancing for allocating comparison tasks, without careful consideration of data distribution and storage usage. While Hadoop-based solutions provide users with simplicity of implementation, their inherent MapReduce computing pattern does not match the ATAC pattern. This leads to load imbalances and poor data locality when Hadoop’s data distribution strategy is used for ATAC problems. Here we present a data distribution strategy which considers data locality, load balancing and storage savings for ATAC computing problems in homogeneous distributed systems. A simulated annealing algorithm is developed for data distribution and task scheduling. Experimental results show a significant performance improvement for our approach over Hadoop-based solutions.

Cognitive workload discrimination of pilots during flights can contribute to flight safety by preventing mental overloading of the aircraft crew. Research has been conducted to study the cognitive workload of pilots in flight simulation task. The estimation of the cortical connectivity is a critical step in mental workload assessment. Therefore, we adopted a novel, parameter-free method of evaluating the cortical connectivity, named Generalized Measure of Association (GMA), to assess and discriminate the mental workload of pilots using the Multi-Attribute Task Battery (MATB) flight simulation platform. A modified version of GMA (Time Series GMA) is applied on the pre-processed EEG time series recorded from eight subjects during the MATB experiment. Frobenius Norm is used to calculate the Euclidean distances between different TGMA series in order to assess the discriminability of the cognitive workloads. The results have shown clear distinction between the mean values of the inter-task and intra-task Euclidean distance series of TGMA matrices, but statistical significance is still lacking due to the relatively large standard deviations.