Search

Highlights • We train a Fully Convolutional Network (FCN) for text prediction in scene images and fuse it with a text proposal method. • Significantly higher recall rates than SoA text localization pipelines and better quality regions are obtained. • The resulting pipeline reduces the number of proposals resulting to a 4 × speed up compared with the baseline. • Our proposed method yields top performance when integrated in an end-to-end pipeline. • Analysis and results on standard datasets COCO-Text and ICDAR-Challenge 4 are reported. Graphical abstract This paper proposes a fusion of a text proposal technique with Fully Convolutional Networks to efficiently reduce the number of proposals while maintaining the same text recall level and thus gaining a significant speed up.Different fusion strategies are explored, as shown in the figure, that demonstrate that such an approach can yield significantly higher recall rates than state-of-the-art text localization techniques, while also producing better-quality localizations. End-to-end performance shows that this recall margin leads to state-of-the-art results in scene text reading systems. Abstract Class-specific text proposal algorithms can efficiently reduce the search space for possible text object locations in an image. In this paper we combine the Text Proposals algorithm with Fully Convolutional Networks to efficiently reduce the number of proposals while maintaining the same recall level and thus gaining a significant speed up. Our experiments demonstrate that such text proposal approaches yield significantly higher recall rates than state-of-the-art text localization techniques, while also producing better-quality localizations. Our results on the ICDAR 2015 Robust Reading Competition (Challenge 4) and the COCO-text datasets show that, when combined with strong word classifiers, this recall margin leads to state-of-the-art results in end-to-end scene text recognition. [ABSTRACT FROM AUTHOR]

Abstract With the advent of cheap smart sensors installed on board marine vehicles and the increasing computational power of small embedded processors there is tremendous potential for the implementation of new strategies to control marine systems on the basis of input-output plant data. The emerging field of smart sensors affords a unique opportunity to have access to on-line measurement of dynamical systems' variables seamlessly, at a low price. By applying a data-driven control algorithm to a marine vehicle, the paper introduces a new perspective on how data can be used in the control loop in marine systems. Classical control methodologies start by developing a model of the plant to be controlled, after which a number of control design techniques can be used. Recent advances in so-called model-free data-driven control methodologies, in particular unfalsified control, hold promise to merge the identification and control phases. Unfalsified control techniques build on the construction of a bank of controllers for a given plant, in which there exists at least one controller that meets the desired performance specification and a falsification unit. The latter is implemented using a cost function that directly evaluates the performance of the controllers (in and out of the feedback loop) using measured input and output data. At each sampling time, the performance of the controllers is assessed and the controllers that do not meet the pre-defined performance specification criteria will be falsified and removed from the bank of the controllers, after which an active controller will be selected among the unfalsified ones. In this paper, by presenting the results of the application of unfalsified control to the problem of Dynamic Positioning (DP) of marine vessels subjected to environmental forces, we aim to attract the attention of researchers in the field of marine control to the new perspective of using data to directly control marine system. [ABSTRACT FROM AUTHOR]

In this paper it is presented a new open source tool for evaluating data fusion systems, mainly related to the maritime surveillance domain. This tool provides specific utilities for designing and simulating synthetic maritime environments to assists in the fusion development process, like designing vessels trajectories, placing different types of sensors, simulating vessels dynamics, or simulating sensors detections. This synthetic information can be used to feed a data fusion system to evaluate its response in a reproducible way under different conditions. This tool can be used to optimize the data fusion evaluation process, since testing a fusion system is a quite complex task, as the fusion output depends on the combination of multiple algorithms, configurations, measures, timing, and so on. Then, having reproducible synthetic environments can be quite useful when evaluating fusion results, system performance in dense scenarios, vessel trajectories with different dynamics, sensor coverages, and so on. This tool has been used with success for evaluating different fusion systems, and now it is presented as an open source tool, so it can be easily adapted to different environments, be used by other researchers, or extended by the community. This paper presents how it is built, the underlying algorithms, and presents some example use cases. [ABSTRACT FROM AUTHOR]

Abstract Community detection is an important task with great practical significance for understanding the structure and function of complex networks in various fields. As the real world networks become larger and more complex, it is a challenge to achieve high quality of community partitioning. In order to identify community structure more effectively in complex networks, a new algorithm, which iteratively deletes edges with restrictions is proposed in this paper. The algorithm first makes use of the connection strength between vertices to divide the original network into some strongly connected communities with optimal modularity by the improved edge-deleting process, and finally reconnects the isolated vertices to initial communities for optimizing community structure. Experiments on the real-world and synthetic networks prove that the proposed algorithm achieves a competitive performance compared with other reference algorithms. [ABSTRACT FROM AUTHOR]

Graphical abstract Highlights • A novel AP-DEACM method is proposed. • Performance evaluation and energy optimization model is obtained. • This proposed model is efficient in energy saving and carbon emission reduction. • The energy saving potential of ethylene and PTA production plants are 2.78% and 1.26%. • Carbon emission reduction potential is 3.62% in ethylene production plants. Abstract Data Envelopment Analysis (DEA) has been widely used in performance and energy efficiency evaluation. However, in the traditional DEA, the effective of each decision making unit (DMU) is evaluated through its own optimized perspective and regardless of other DMUs influence, which may result in too many effective DMUs. And the DEA cross-model (DEACM) can distinguish the effective DMUs better by constructing a cross-efficiency matrix, but the optimal weight of the DMU may not be unique, so the cross efficiency of the DEACM may be different. Therefore, this paper proposes a novel DEACM based on the affinity propagation (AP) clustering algorithm (AP-DEACM). Through the AP clustering algorithm, the high impact data affecting the performance capacity and energy saving are obtained. Then the better effective DMU is identified through the high discrimination of the AP–DEACM. Finally, the proposed AP-DEACM is used for performance evaluation and energy optimization modeling of the Pure Terephthalic Acid (PTA) production process and the ethylene industrial process in complex petrochemical industries. The experimental results show that the energy saving potential of PTA production plants and ethylene production plants are 2.78% and 1.26%, respectively, and the average value of carbon emission savings potential is 3.62% in ethylene production plants. [ABSTRACT FROM AUTHOR]

Highlights • A practical path planning algorithm has been proposed for USV navigation. • A new path smoothing process is developed to improve the quality of the path. • The algorithm is integrated with a real USV to achieve the autonomous navigation. • Both field trials and computer simulations were conducted to verify the algorithm. Abstract An effective path planning or route planning algorithm is essential for guiding unmanned surface vehicles (USVs) between way points or along a trajectory. The A* algorithm is one of the most efficient algorithms for calculating a safe route with the shortest distance cost. However, the route generated by the conventional A* algorithm is constrained by the resolution of the map and it may not be compatible with the non-holonomic constraint of the USV. In this paper an improved A* algorithm has been proposed and applied to the Springer USV. A new path smoothing process with three path smoothers has been developed to improve the performance of the generated route, reducing unnecessary 'jags', having no redundant waypoints and offering a more continuous route. Both simulation and experimental results show that the smoothed A* algorithm outperforms the conventional algorithm in both sparse and cluttered environments that have been uniformly rasterised. It has been demonstrated that the proposed improved A* route planning algorithm can be applied to the Springer USV providing promising results when tracking trajectories. [ABSTRACT FROM AUTHOR]

Timing-constrained power-driven gate sizing has aroused lot of research interest after the recent discrete gate sizing contests organized by International Symposium on Physical Design. Since then, there are plenty of research papers published and new algorithms are proposed based on the contest formulation. However, almost all (new and old) papers in the literature ignore the details of how power-driven gate sizing fits in industrial physical synthesis flows, which limits their practical usage. This paper aims at filling this knowledge gap. We explain our approach to integrate a state-of-the-art Lagrangian Relaxation-based gate sizing into our actual physical synthesis framework, and explain the challenges and issues we observed from the point of view of VLSI design flows. [ABSTRACT FROM AUTHOR]

Probabilistic K -nearest neighbour (PKNN) classification has been introduced to improve the performance of the original K -nearest neighbour (KNN) classification algorithm by explicitly modelling uncertainty in the classification of each feature vector. However, an issue common to both KNN and PKNN is to select the optimal number of neighbours, K . The contribution of this paper is to incorporate the uncertainty in K into the decision making, and consequently to provide improved classification with Bayesian model averaging. Indeed the problem of assessing the uncertainty in K can be viewed as one of statistical model selection which is one of the most important technical issues in the statistics and machine learning domain. In this paper, we develop a new functional approximation algorithm to reconstruct the density of the model (order) without relying on time consuming Monte Carlo simulations. In addition, the algorithms avoid cross validation by adopting Bayesian framework. The performance of the proposed approaches is evaluated on several real experimental datasets. [ABSTRACT FROM AUTHOR]

Abstract: In a wireless sensor network (WSN), data collected by a sensor node need to be associated with location information in order to support real-world applications. Taking the WSN characteristics into account, this paper proposes an address configuration scheme based on location information and passive duplicate address detection (PDAD). In this scheme, the network architecture based on location information is presented, and based on this architecture the address initialization algorithm and address maintenance algorithm are proposed. The address initialization algorithm is performed once the network starts, and it is made up of the initialization address configuration sub-algorithm (IAC) and the initialization location PDAD sub-algorithm (ILPDAD). The address maintenance algorithm is performed after the initialization algorithm is complete, and it is composed of the maintenance address configuration sub-algorithm (MAC) and the maintenance location PDAD sub-algorithm (MLPDAD). During the address initialization process (address maintenance process), a new node first uses IAC (MAC) to obtain an address and then performs ILPDAD (MLPDAD) to ensure the address uniqueness. Since beacon frames are employed to achieve IAC (MAC) and ILPDAD (MLPDAD), the address configuration cost and delay are reduced. Moreover, IAC (MAC) is based on location information and ILPDAD (MLPDAD) is based on PDAD, so there are always the sufficient address resources for address configuration without address reclamation. In this way, the extra cost and delay caused by both the address reclamation and the address configuration failure are avoided. This paper evaluates the performance of this scheme, and the data results show that this scheme effectively reduces the address configuration delay and cost. [Copyright &y& Elsevier]

Highlights: [•] The paper has solved different Economic Emission Load Dispatch (EELD) problems. [•] The paper considered both single and multi-objective emission dispatch problems. [•] Real Coded Chemical Reaction Optimization (RCCRO) is applied to solve the problem. [•] Performance of RCCRO has been compared with BBO, modified BFA and GSA. [•] Performance of RCCRO has been found to be more encouraging to solve EELD problems. [ABSTRACT FROM AUTHOR]

In this paper, we analyze applicability of various load-balancing methods in countering data skew in MapReduce computations. A MapReduce job consists of several phases: mapping, shuffling data, sorting and reducing. The distribution of the work in the last three phases is data-driven, and unequal distribution of the data keys may cause imbalance in the computation completion times and prolonged execution of the whole job. We propose algorithms of four different types for balancing computational effort in reduce-heavy MapReduce jobs and evaluate their performance under various degrees of data skew and system parameters. By applying an innovative method of visualizing algorithm dominance conditions, we are able to determine conditions under which certain load-balancing algorithms are capable of scheduling MapReduce computations well. We conclude that no single algorithm is a panacea and hybrid approaches are necessary. [ABSTRACT FROM AUTHOR]

Visemes are the visual equivalent of phonemes. Although not precisely defined, a common working definition of a viseme is “a set of phonemes which have identical appearance on the lips”. Therefore a phoneme falls into one viseme class but a viseme may represent many phonemes: a one-to-many mapping. This mapping introduces ambiguity between phonemes when using viseme classifiers. Not only is this ambiguity damaging to the performance of audio-visual classifiers operating on real expressive speech, there is also considerable choice between possible mappings. In this paper we explore the issue of this choice of viseme-to-phoneme map. We show that there is definite difference in performance between viseme-to-phoneme mappings and explore why some maps appear to work better than others. We also devise a new algorithm for constructing phoneme-to-viseme mappings from labeled speech data. These new visemes, ‘Bear’ visemes, are shown to perform better than previously known units. [ABSTRACT FROM AUTHOR]

Multi-instance learning (MIL) is commonly used to classify a set of instances, also known as a bag, where labels for the training set are only available for each bag. Many MIL methods exist, but they often suffer from high computation complexity and the key information from MIL being ignored, which deteriorates the classification performance. Recently, locality-sensitive hashing (LSH), with its high scalability, has shown the ability in enhancing MIL performance. However, for these LSH-based methods, the fixed number of bits is used to represent each projected dimension, resulting in subtle information loss and the algorithm performance reduction. In this paper, we propose a self-adaptive LSH encoding method for MIL, termed as SALE. SALE uses LSH to generate the primary batches, followed by a self-adaptive process for reconstruction. Reconstructed bags are transformed into random super histograms (RSH) using an incomplete coding method, and then weighted through a scheme that takes advantage of key instances. These weighted RSHs are used to train the learning model. SALE efficiently deals with large MIL problems, due to its low complexity and RSH’s ability to exploit key information of MIL. Experiments demonstrate SALE’s good performance compared to state-of-the-art MIL methods. [ABSTRACT FROM AUTHOR]

Seam carving, also known as content-aware image resizing, is the most popular image resizing algorithm nowadays. Therefore, detecting seam carving has become an important topic in image forensics. In this paper, an advanced statistical model, consisting of local derivative pattern, Markov transition probabilities, and subtractive pixel adjacency model, is proposed to determine if an image has been seam carved or not. The performance of the proposed feature set can be further improved, and the feature set's dimensionality can be largely reduced by utilizing linear support vector machine (SVM) based recursive feature elimination. With the linear SVM classifier, the experimental works have demonstrated that the proposed approach can successfully detect seam carving. It outperforms the state-of-the-art in general; in particular at the low carving rate cases, such as 5%, 10% and 20%, the average detection accuracy has been boosted from 66%, 75% and 87% to 81%, 90% and 96%, respectively. On detecting seam carving in JPEG images and geometrical transformed uncompressed images, the proposed approach has also shown promising performance. [ABSTRACT FROM AUTHOR]

The design of a Homomorphic Encryption (HE) algorithm that allows computation over the encrypted data is required in real world modern applications to preserve the privacy. Such applications include Cloud Computing, shared storage, processing resources, etc. The existing solutions are not practical for real world applications. Asymmetric approaches suffer from high computation overhead, while symmetric approaches suffer from low immunity against attacks such as chosen and known plaintext attack. In this paper, we consider symmetric approaches and we focus on Matrix Operation for Randomization and Encryption (MORE) approach to build a new algorithm overcoming the drawbacks of MORE. The proposed algorithm is explained in details and evaluated. The security performance results show that the proposed approach can prevent the strong attacks without degradation of the system performances in term of latency and energy consumption. [ABSTRACT FROM AUTHOR]

This paper primarily proposes a novel method, named maximum correlation information (MCI), to evaluate the importance of each feature by maximizing correlation information between feature space and class coding space. Then the proposed MCI is combined with the strategy of recursive feature elimination to form a new feature selection method, MCI-based recursive feature elimination (MCI-RFE). MCI-RFE aims to select the optimal features with lower time complexity. To validate the performance of the proposed method, random 10-fold cross-validation is applied thirty times on six widely used benchmark datasets with integrated features from position-specific score matrix (PSSM), PROFEAT and Gene Ontology (GO). The experimental results show that MCI-RFE is highly competitive and works effectively for integrated high-dimensional protein data compared to the other state-of-the-art algorithms including SVM-RFE, ReliefF-RFE and Random-Forest. [ABSTRACT FROM AUTHOR]

The acceleration autopilot with a rate loop is the most commonly implemented autopilot, which has been extensively applied to high-performance missiles. Nevertheless, for spinning rockets, the design of the guidance and control modules is a challenging task because the rapid spinning of the body creates a heavy coupling between the normal and lateral rocket dynamics. Nonlinear modeling of the rocket dynamics, control design as well as guidance algorithms are performed in this paper. Moreover, discrete-time guidance and control algorithms for the terminal phase, which is based on proportional navigation, are performed. Finally, complete nonlinear simulations based on realistic scenarios are developed to demonstrate the robustness of the proposed solution with respect to uncertainty regarding launch, environment and rocket conditions. The performance of the proposed navigation, guidance and control system for a high-spin rocket leads to significant reductions in impact point dispersion. [ABSTRACT FROM AUTHOR]

Through decreasing inter-cell interferences, CoMP (Coordinated Multi-Point) can enhance system capacity and spectral efficiency of cell-edge users significantly. BSs (Base stations) are the front-end accesses of CoMP and their planning influences the overall performance of the system directly. Different from the traditional cellular network, the coverage of BSs in CoMP always overlaps with each other to some degree. In order to investigate the BSs planning for CoMP, this paper builds two optimization models to obtain the location and the number of the BSs. A joint optimization algorithm is proposed to solve models, in which the methodologies of mesh adaptive direct search and successive elimination have been used. Numerical results show that the proposed models can effectively reduce the total transmission power and the number of BSs without losing the quality of service for users, which verifies the effectiveness of the proposed algorithms and the inherent advantage of CoMP. [ABSTRACT FROM AUTHOR]

Mobile sink (MS) has drawn significant attention for solving hot spot problem (also known as energy hole problem) that results from multihop data collection using static sink in wireless sensor networks (WSNs). MS is regarded as a potential solution towards this problem as it significantly reduces energy consumption of the sensor nodes and thus enhances network lifetime. In this paper, we first propose an algorithm for designing efficient trajectory for MS, based on rendezvous points (RPs). We next propose another algorithm for the same problem which considers delay bound path formation of the MS. Both the algorithms use k -means clustering and a weight function by considering several network parameters for efficient selection of the RPs by ensuring the coverage of the entire network. We also propose an MS scheduling technique for effective data gathering. The effectiveness of the proposed algorithms is demonstrated through rigorous simulations and comparisons with some of the existing algorithms over several performance metrics. [ABSTRACT FROM AUTHOR]

The number of methods available for classification of multi-label data has increased rapidly over recent years, yet relatively few links have been made with the related task of classification of sequential data. If labels indices are considered as time indices, the problems can often be seen as equivalent. In this paper we detect and elaborate on connections between multi-label methods and Markovian models, and study the suitability of multi-label methods for prediction in sequential data. From this study we draw upon the most suitable techniques from the area and develop two novel competitive approaches which can be applied to either kind of data. We carry out an empirical evaluation investigating performance on real-world sequential-prediction tasks: electricity demand, and route prediction. As well as showing that several popular multi-label algorithms are in fact easily applicable to sequencing tasks, our novel approaches, which benefit from a unified view of these areas, prove very competitive against established methods. [ABSTRACT FROM AUTHOR]

Intel's Advanced Vector Extensions is known as single instruction multiple data streams (SIMD), and the instruction sets is introduced in the second-generation Intel Core processor family. This new technology is supported by new generations of Intel and AMD processors . The advanced vector extensions (AVX) exploits single instruction multiple data (SIMD) computing units for fine grained-parallelism. These instructions process multiple data elements simultaneously and independently. Many applications such as signal processing, recognition, visual processing, scientific and engineering numerical, physics and other areas of applications need for vector floating point performance supported by AVX. Matrix-Matrix multiplications is the core of many important algorithms such as signal processing, scientific and engineering numerical, so it is substantial to accelerate implementation of matrix-matrix multiplications. It is very important to use appropriate compilers that can optimally utilize the new features of the evolving processors. For this purpose, a clear vision on the performance of the compilers on performance characteristics of AVX is needed. In addition choosing the appropriate programming method is substantial to gain the best performance. In this paper, the performance evaluation of matrix-matrix multiplications in three forms (C = A. B, C = A. B T , and C = A T . B), using Intel's advanced vector extension (AVX) instruction sets has been reported. The obtained results are compared using inline assembly versus intrinsic functions for programming. A comparative study to indicate the effects of two widely used C++ compilers: Intel C++ compiler (ICC) in Intel Parallel Studio XE 2016 against Microsoft Visual Studio C++ compiler 2015 (MSVC++) has been investigated. The results are evaluated on Intel Core i7 processor on a Broadwell system for square matrices of different large sizes. The results demonstrate that the Intel compiler has better performance than MSVC++ compiler by 1.34, 1.32, and 1.22 using inline assembly language and by 1.36, 1.19, and 1.25 using intrinsic functions for C = A. B, C = A. B T , and C = A T . B, respectively. The performance of using intrinsic functions compared to the inline assembly demonstrates that the intrinsic functions has better performance than inline assembly by 2.1, 2.13, and 2.18 using Intel compiler and by 2.08, 2.49, and 2.11 using MSVC++ compiler for C = A. B, C = A. B T , and C = A T . B, respectively. [ABSTRACT FROM AUTHOR]

Numerous different methodologies have been introduced in the last few decades to provide efficient solutions for complex real-world problems and other optimization problems. This work focuses on the development of a simple hybrid cultural learning theme with a balanced performance for differential evolution frameworks. It is intended to be always efficient for a diverse set of optimization tasks. As different optimization algorithms behave differently depending on the problems, the combination of the best behaviors from different search strategies seems desirable. The proposed work explores the combination of the explorative/exploitative strengths of two heuristic search techniques, which discretely provide competitive results. Differential evolution is used as the population space for Cultural Algorithm, and is used to guide knowledge dissemination from the knowledge sources in the belief space. Here, a new influence function is introduced that adjusts the membership of each of the knowledge sources. The algorithm has been tested with the conditions and benchmark problems defined for the IEEE CEC2013 special session and competition on real-parameter single objective optimization. The paper also investigates the application of the new algorithm to a set of real-life problems concerning optimizing the weight a tension/compression spring and minimizing the fabrication cost of a welded beam engineering problem. The proposed algorithm appears to have a significant impact on the algorithmic functioning as it reliably augments the performance of the differential evolution frameworks with which it is integrated. Benchmark results for most of the synthetic functions from the special session show that the balanced hybrid obtains superior performance compared to the other competent algorithms. It scales well with the increasing dimensionality and converges in the close proximity of the global optimum for complex functions. [ABSTRACT FROM AUTHOR]

As a novel overlapping community detection theory, topology potential has inspired many methods. However, these methods ignore the mass difference between nodes, leading to inaccurate topological potential values of nodes. Moreover, additional strategies are needed to determine the community affiliation of nodes, further complicating the process of community detection. In this paper, we propose a new overlapping community detection method based on node location analysis. In the proposed method, the PageRank algorithm is used to evaluate the node mass, and the community affiliation of nodes is determined based on their positions in the inherent peak-valley structure of the topology potential field. Experimental results show that the proposed method exhibits excellent performance on artificial and real-world networks and outperforms other topology-potential-based and most non-topology-potential-based methods. [ABSTRACT FROM AUTHOR]

Due to the rapid development of application for medical images in recent years, the security of medical images becomes more vital. In this paper, a fragile watermarking scheme for medical images is proposed. The proposed scheme first divides the whole medical image into ROI (Regions of Interest) and RONI (Regions of Non-interest). Then the JPEG-LS algorithm is applied to ROI to get the compressed data which will be embedded into the whole medical image. And then, image encryption methods are used to assure the invisibility of the medical image. At last, a POB number system enhances the security of the scheme. With the use of the scheme in this paper, we can judge whether the medical image is tampered or not, and can also recover the original medical images in specific conditions without any loss. Moreover, if the medical image is tampered, we can also localize the tampered regions. Experimental results show that the suggested scheme achieves a better performance on security, invisibility and robustness in comparison to the state-of-the-art methods for medical images. [ABSTRACT FROM AUTHOR]

Detection of community structures is beneficial to analyzing the structures and properties of networks. It is of theoretical interest and practical significance in modern science. So far, a large number of algorithms have been proposed to detect community structures in complex networks, but most of them are suitable for a specific network structure. In this paper, a novel method (called CDMIC) is proposed to detect the communities in un-weighted, weighted, un-directed, directed and signed networks by constructing a dissimilarity distance matrix of network and identifying community centers with maximizing modularity. For a given network, we first estimate the distance between all pairs of nodes for constructing the dissimilarity distance matrix of the network. Then, this distance matrix is input to the affinity propagation (AP) algorithm to extract a candidate center set of community. Thirdly, we rank these centers in descending order according to the sum of their availability and responsibility. Finally, we determine the community structure by selecting the center subset from the candidate center set in an incremental manner to make the modularity maximization. On three real-world networks and some synthetic networks, experimental results show that our CDMIC method has higher performance in terms of classification accuracy and normalized mutual information (NMI), and ability to tolerate the resolution limitation. [ABSTRACT FROM AUTHOR]

In the basic setting of formal concept analysis, a many-valued attribute needs to be replaced with several one-valued attributes. These one-valued attributes can be interpreted as a certain level of granularity of the corresponding many-valued attribute. In this paper, we explore theoretical relationships between concepts before and after increasing the granularity level of one attribute, based on which we introduce an efficient method of concept classification. Moreover, a new preprocessing routine is proposed to help generate new concepts and restore lattice order relation. These two procedures can considerably reduce the comparisons between sets, compared to the original Zoom-In algorithm. By employing these two procedures, we introduce an efficient algorithm, referred to as Unfold, to increase the granularity levels of attributes. The algorithm can perform a Zoom-In operation on a concept lattice associated with a coarser data granularity to obtain a new one that consists of finer formal concepts without building the new lattice from scratch. We describe the algorithm and present an experimental evaluation of its performance and comparison with another Zoom-In algorithm. Empirical analyses demonstrate that our algorithm is superior when applied to various types of datasets. [ABSTRACT FROM AUTHOR]

Very recently, a new metaheuristic called artificial raindrop algorithm (ARA) was proposed. This search algorithm inspired from the phenomenon of natural rainfall, including the generation of raindrop, the descent of raindrop, the collision of raindrop, the flowing of raindrop and the updating of vapor. However, the original ARA only focused on the changing process of a raindrop. In this paper, we present an extension of ARA ( ARA E ) to more raindrops without any major conceptual change to its structure. In the proposed ARA E , all vapors are dynamically divided into several small-sized groups according to the relative distance of vapors in each generation. Each vapor has an associated weight proportion to the fitness value. The raindrop in the corresponding group is then generated based on weighted mean of the current vapors positions. But beyond that, some operators of ARA are further modified to enhance its exploration/exploitation capabilities. In order to thoroughly evaluate the performance of ARA E , a comprehensive comparative study has been carried on the CEC2005 contest benchmark functions. The obtained results indicate that ARA E has overall better performance than ARA, and is very competitive with respect to other twenty-four state-of-the-art original intelligent optimization algorithms and twenty-four improved metaheuristic algorithms. Finally, the proposed ARA E is also applied to artificial neural networks and the promising results on the nonlinear function approximation show the applicability of ARA E for problem solving. [ABSTRACT FROM AUTHOR]

The traditional manifold learning methods can preserve the local sub-manifold structure or the global geometry effectively, such as elastic preserving projections (EPP). Many experimental results have been shown that EPP, a recently developed linear algorithm, is a strong analyzer for high-dimensional data. However, for classification problems, the traditional methods focused on the geometrical information and ignores discriminative information of different classes. In this paper, we propose a novel discriminative orthogonal elastic preserving projections (DOEPP) by imposing the discriminant information and the orthogonal constraint to improve its classification performance. DOEPP does not only preserve the elasticity of the training set, but also sufficiently utilizes the discriminant information by adding maximum margin criterion and the orthogonality of the projection matrix into its objective function. Extensive experiments on two well-known synthetic manifold data sets and four publicly available databases illustrate the effectiveness of our method. [ABSTRACT FROM AUTHOR]

Occlusion in face recognition is a common yet challenging problem. While sparse representation based classification (SRC) has been shown promising performance in laboratory conditions (i.e. noiseless or random pixel corrupted), it performs much worse in practical scenarios. In this paper, we consider the practical face recognition problem, where the occlusions are predictable and available for sampling. We propose the structured occlusion coding (SOC) to address occlusion problems. The structured coding here lies in two folds. On one hand, we employ a structured dictionary for recognition. On the other hand, we propose to use the structured sparsity in this formulation. Specifically, SOC simultaneously separates the occlusion and classifies the image. In this way, the problem of recognizing an occluded image is turned into seeking a structured sparse solution on occlusion-appended dictionary. In order to construct a well-performing occlusion dictionary, we propose an occlusion mask estimating technique via locality constrained dictionary (LCD), showing striking improvement in occlusion sample. On a category-specific occlusion dictionary, we replace l 1 norm sparsity with the structured sparsity which is shown more robust, further enhancing the robustness of our approach. Moreover, SOC achieves significant improvement in handling large occlusion in real world. Extensive experiments are conducted on public data sets to validate the superiority of the proposed algorithm. [ABSTRACT FROM AUTHOR]

Krill herd (KH) has been proven to be an efficient algorithm for function optimization. For some complex functions, this algorithm may have problems with convergence or being trapped in local minima. To cope with these issues, this paper presents an improved KH-based algorithm, called Opposition Krill Herd (OKH). The proposed approach utilizes opposition-based learning (OBL), position clamping (PC) and Cauchy mutation (CM) to enhance the performance of basic KH. OBL accelerates the convergence of the method while both PC and heavy-tailed CM help KH escape from local optima. Simulations are implemented on an array of benchmark functions and two engineering optimization problems. The results show that OKH has a good performance on majority of the considered functions and two engineering cases. The influence of each individual strategy (OBL, CM and PC) on KH is verified through 25 benchmarks. The results show that the KH with OBL, CM and PC operators, has the best performance among different variants of OKH. [ABSTRACT FROM AUTHOR]

The importance of metrics in machine learning and pattern recognition algorithms has led to an increasing interest for optimizing distance metrics in recent years. Most of the state-of-the-art methods focus on learning Mahalanobis distances and the learned metrics are in turn heavily used for the nearest neighbor-based classification (NN). However, until now no theoretical link has been established between the learned metrics and their performance in NN. Although some existing methods such as large-margin nearest neighbor (LMNN), have employed the concept of large margin to learn a data-dependent metric, the link between the margin and the generalization performance for the metric is not fully understood. Though the recent work has indeed provided tenable margin distribution explanation on Boosting, the margin used in metric learning is quite different from that in Boosting. Thus, in this paper we try to analyze the effectiveness of metric learning algorithms for NN from the perspective of the margin distribution and provide a general and effective evaluation criterion for metric learning. On the one hand, we derive the generalization error upper bound for NN with respect to the Mahalanobis metric. On the other hand, the experiments on several benchmark datasets using existing metric learning algorithms demonstrate that large margin distribution can be obtained by these algorithms. Motivated by our analysis above, we also present a novel margin based metric learning algorithm for NN, which explicitly enlarges the margin distribution on various datasets and achieves very competitive results with the existing metric learning algorithms. [ABSTRACT FROM AUTHOR]

This paper presents a pre-processing and a distance which improve the performance of machine learning algorithms working on independent and identically distributed stochastic processes. We introduce a novel non-parametric approach to represent random variables which splits apart dependency and distribution without losing any information. We also propound an associated metric leveraging this representation and its statistical estimate. Besides experiments on synthetic datasets, the benefits of our contribution is illustrated through the example of clustering financial time series, for instance prices from the credit default swaps market. Results are available on the website http://www.datagrapple.com and an IPython Notebook tutorial is available at http://www.datagrapple.com/Tech for reproducible research. [ABSTRACT FROM AUTHOR]

With the rapid advances in 3D capture and display technology, tag ranking for stereo images will be a potential application on web image retrieval. Directly extending the existing approaches to stereo images is problematic as it may neglect the representative 3D elements. In this paper, a novel automatic tag saliency ranking algorithm for stereo images is presented. Specifically, a novel method of interacting with stereo images is proposed to segment the two images into regions simultaneously. Then tags annotated on the image-level are propagated to the region-level via an improved multi-instance learning algorithm. In the next, a new 3D saliency detection algorithm is proposed using occlusion cues along with the contrast in color and disparity. And finally, tags are re-ranked according to the saliency values of the corresponding regions. Moreover, to evaluate the performance of our approach, an annotated stereo image dataset is set up. The experimental results on this dataset demonstrate the effectiveness of our approach. [ABSTRACT FROM AUTHOR]

Different methods have been used to train radial basis function (RBF) neural networks. This paper proposes the use of a bee-inspired algorithm, named cOptBees, plus a heuristic to automatically select the number, location and dispersions of basis functions to be used in RBF networks. cOptBees was originally designed to solve data clustering problems and the prototypes determined by the algorithm will be selected as the centers for the RBF network. The presented approach, named BeeRBF, is used to solve classification problems and is evaluated both in terms of the decision boundaries generated and classification accuracy. The performance of BeeRBF was compared with that of k- means, random center selection and some other proposals from the literature. The results show that BeeRBF is competitive and has the advantage of automatically determining the number of centers to be used in the RBF network. [ABSTRACT FROM AUTHOR]

The latest developments in the domain of HPC have lead to the deployment of complex extreme-scale systems, based on diverse computing devices (CPU, GPU, accelerators) thus posing the question of scalability in the light not only of parallel efficiency, but also in terms of energy efficiency. In this paper we propose a new metrics for energy aware performance estimation based on our experience and the analysis of the existing metrics. We study the performance of computationally intensive Monte Carlo applications deployed on heterogeneous HPC systems with focus on energy efficiency and equipment costs. We compare the energy aware performance results of CPU and GPU variants of the tested algorithms with respect to the introduced measures and metrics. The results of our study demonstrate the importance of taking into account not only scalability of the HPC applications but also energy efficiency and equipment cost. They also show how to optimize the selection of CPU computing or computing with GPGPUs. The results can be used by application developers/users and also by resource providers. [ABSTRACT FROM AUTHOR]

In this paper, a novel algorithm named Semi-supervised Local Ridge Regression (SSLRR) is proposed for local matching based face recognition. Compared with other algorithms, the proposed algorithm possesses two advantages. Firstly, SSLRR utilizes a multiple graph based semi-supervised technique to propagate the class labels of labeled samples to the unlabeled ones. Thus, the information of both labeled and unlabeled data can be employed in our algorithm to improve its performance. Secondly, unlike most local matching based face recognition algorithms which assume different sub-images from the same face are independent, an adaptive non-negative weight vector is introduced into our SSLRR to combine the Laplacian matrices obtained by different sub-images. Therefore, the latent complementary information of multiple sub-patterns from the same face image can be taken into account. Moreover, a simple yet efficient iterative update scheme is also proposed to solve our SSLRR model. Extensive experiments are performed on five standard face databases (Yale, Extended YaleB, AR, CMU PIE and LFW) to demonstrate the efficiency of the proposed algorithm. Experimental results show that SSLRR obtains better recognition performance than some other state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

Co-training is one of the major semi-supervised learning paradigms in which two classifiers are alternately trained on two distinct views and they teach each other by adding the predictions of unlabeled data to the training set of the other view. Co-training can achieve promising performance, especially when there is only a small number of labeled data. Hence, co-training has received considerable attention, and many variant co-training algorithms have been developed. It is essential and informative to provide a systematic framework for a better understanding of the common properties and differences in these algorithms. In this paper, we propose a general framework for co-training according to the diverse learners constructed in co-training. Specifically, we provide three types of co-training implementations, including co-training on multiple views, co-training on multiple classifiers, and co-training on multiple manifolds. Finally, comprehensive experiments of different methods are conducted on the UCF-iPhone dataset for human action recognition and the USAA dataset for social activity recognition. The experimental results demonstrate the effectiveness of the proposed solutions. [ABSTRACT FROM AUTHOR]

This paper presents a meta-objective optimization approach, called Bi-Goal Evolution (BiGE), to deal with multi-objective optimization problems with many objectives. In multi-objective optimization, it is generally observed that 1) the conflict between the proximity and diversity requirements is aggravated with the increase of the number of objectives and 2) the Pareto dominance loses its effectiveness for a high-dimensional space but works well on a low-dimensional space. Inspired by these two observations, BiGE converts a given multi-objective optimization problem into a bi-goal (objective) optimization problem regarding proximity and diversity, and then handles it using the Pareto dominance relation in this bi-goal domain. Implemented with estimation methods of individuals' performance and the classic Pareto nondominated sorting procedure, BiGE divides individuals into different nondominated layers and attempts to put well-converged and well-distributed individuals into the first few layers. From a series of extensive experiments on four groups of well-defined continuous and combinatorial optimization problems with 5, 10 and 15 objectives, BiGE has been found to be very competitive against five state-of-the-art algorithms in balancing proximity and diversity. The proposed approach is the first step towards a new way of addressing many-objective problems as well as indicating several important issues for future development of this type of algorithms. [ABSTRACT FROM AUTHOR]

Non-parametric Nearest-Neighbour (NN) image classification is desired in certain applications, because no intensive learning is required. Naive Bayes Nearest Neighbour (NBNN) and its improved version, Local Naive Bayes Nearest Neighbour (Local NBNN), are two impressive classifiers that keep a good balance between algorithm accuracy and complexity. Instead of computing image-to-image (I2I) distances, these two algorithms calculate image-to-class (I2C) distances. As a consequence, the local image descriptors are not quantised and the performance of such an image classifier is thereby enhanced. In this paper, by applying the concept of saliency detection to the calculation of I2C distances, we separate each training image from each category into two parts: foreground and background. Then we calculate I2C distances for foreground and background to classify the test image. Afterwards, the scores presented by the foreground and the background are used to correct the responses generated by the original NBNN or Local NBNN. Moreover, it is observed that a prior clustering inside each training category is able to reduce time consumption significantly without sacrificing system performance. By combining the contributions above, our approach is superior to the original NBNN and Local NBNN classifiers in terms of both efficiency and accuracy on three datasets: Pami-09, 15-Scene and Caltech-5. [ABSTRACT FROM AUTHOR]

It is well known that in supervised learning, active learning could effectively decrease the complexity of training instances without obvious loss of the classification performance. Generally, active learning is applied in the scenario that lots of instances are easy to be acquired, but labeling them is expensive and/or time-consuming. In this study, we try to implement active learning by using extreme learning machine (ELM) classifier based on three reasons as follows: (1) ELM has light computational costs, (2) ELM has strong generalization ability which is even comparable with support vector machine (SVM) and (3) ELM could be directly applied on both binary-class and multiclass problems. Specifically, an active learning algorithm based on ELM classifier named AL-ELM is proposed in this paper. During active learning, AL-ELM estimates the uncertainty of each unlabeled instance by creating a mapping relation between the actual outputs of the instance in ELM and the approximated membership probability of the same instance. In other words, ELM is converted as the equivalent Bayes classifier. On each iteration, those most uncertain instances are extracted and labeled to promote the quality of classification model. The learning procedure stops until it satisfies a pre-designed criterion. Experimental results on 20 benchmark data sets show that AL-ELM is better than or at least comparable to several state-of-the-art uncertainty-based active learning algorithms. Also, in contrast with several other algorithms, AL-ELM could effectively decrease the running time of learning procedure. [ABSTRACT FROM AUTHOR]

This paper proposed a wavelet quantization based method for robust watermarking with resistance to incidental distortions. For transform domain based watermarking algorithms, blindly localizing adequate significant coefficients is one critical issue to guarantee the robustness while preserving good fidelity. In the proposed method, low frequency wavelet coefficients of the host image are randomly permutated into sub-groups according to a watermarking secret key. Embedding modifications are then distributed to important coefficients which preserve large perceptual capacity by quantizing the significant amplitude difference (SAD). Meanwhile, dither modulation strategy is employed to control the quantization artifacts and increase the robustness. In such a framework, the blind watermark extraction can be straightforwardly achieved with the watermarking secret keys which only shared by the embedder and extractor for advanced security. Numerous comparison experiments are conducted to evaluate the watermarking performance. Experimental results demonstrate the superiority of our scheme on robustness against content-preserving operations and incidental distortions such as JPEG compression, Gaussian noise. [ABSTRACT FROM AUTHOR]

This paper gives a multi-view relational fuzzy c-medoid vectors clustering algorithm that is able to partition objects taking into account simultaneously several dissimilarity matrices. The aim is to obtain a collaborative role of the different dissimilarity matrices in order to obtain a final consensus fuzzy partition. These matrices could have been obtained using different sets of variables and dissimilarity functions. This algorithm is designed to give a fuzzy partition and a vector of medoids for each fuzzy cluster as well as to learn a relevance weight for each dissimilarity matrix by optimizing an objective function. These relevance weights change at each iteration of the algorithm and are different from one cluster to another. Moreover, various tools for interpreting the fuzzy partition and fuzzy clusters provided by this algorithm are also presented. Several examples illustrate the performance and usefulness of the proposed algorithm. [ABSTRACT FROM AUTHOR]

This paper addresses the nonlinear observer design problem for gene regulatory networks with time-varying delay, focusing on the case of unstable GRNs with oscillatory behavior. Currently available approaches are conservative due to presence of nonlinear terms, which should be dealt with. In addition, nonlinear terms are only known approximately in practice and therefore previous works may lead to undesirable performance. To address conservativeness issue, we have provided an LPV approach. Besides, to diminish effects of uncertain nonlinear terms on observer performance, a peak-to-peak state estimation problem is considered. By defining appropriate Lyapunov–Krasovskii functional, sufficient conditions are derived that guarantee the desirable performance. The superiority of the proposed method to existing approaches is illustrated by means of a numerical example. Moreover, good performance of state estimation in presence of uncertainty is demonstrated by simulations. [ABSTRACT FROM AUTHOR]

The image decomposition based method is one of the efficient and important face recognition solutions for the single sample per person problem. The low image decomposition performance and the unconvincing reconstruction of the approximation image are the two main limitations of the previous methods. In this paper, a new single sample face recognition method based on lower-upper (LU) decomposition algorithm is proposed. The procedure of the proposed method is as following. First, the single sample and its transpose are decomposed to two sets of basis images by using the LU decomposition algorithm, which is more efficient than the image decomposition algorithms of the previous works. Two approximation images are reconstructed from the two basis image sets by the reverse thinking approach based on experimental analysis. Then, the fisher linear discriminant analysis (FLDA) algorithm is used to evaluate the optimal projection space by using the new training set consisting of the single sample and its two approximation images for each person. Finally, the nearest neighbor classifier based on Euclidean distance is adopted as the final classification. We make two main contributions: one is that we propose to decompose the single sample and its transpose using the efficient LU decomposition algorithm, and reorder each basis image set according to the basis image energy; the other is that we present a reverse thinking approach based on experimental analysis to reconstruct the approximation image. The performance of the proposed method is verified using four public face databases, namely FERET, AR, ORL and Yale B. The experimental results indicate that the proposed method is efficient and outperforms several state-of-the-art approaches which are proposed to address the single sample per person problem. [ABSTRACT FROM AUTHOR]

In this paper, we introduce the adaptive natural gradient method into the multilayer perceptrons (MLPs) and radial basis function (RBF) networks. We give a good performance for the Mackey–Glass chaotic time series prediction, and compare it with the LMA. Results show that the adaptive natural gradient methods for MLPs and RBFs, which are the online learning, can give almost the same performance with the LMA (MLPs), which is the batch mode learning. However, the performance of LMA (RBFs) is very poor and is very sensitive with the initial parameters. [ABSTRACT FROM AUTHOR]

The classical AdaBoost algorithm is an ensemble of weak learners, and can be used to construct a strong classifier. A weak learner is incorporated into the ensemble at each step, and the classification of the derived ensemble is improved by properly adjusting the weight of each weak learner. The classical AdaBoost algorithm has some limitations; for example, it is sensitive to noisy data, which may impede the generalization capability of the derived classifier and lead to overfitting problem. A selective boosting, sBoost, technique is proposed in this paper to tackle these problems. The proposed sBoost classifier focuses on the generalized classification performance rather than those hard-to-learn samples, and the penalties of hard-to-learn samples are mitigated to the degree associated with their noise level. An error correction method is suggested to detect potential clean samples and prevent them from misclassification to further alleviate the overfitting problem. The effectiveness of the developed sBoost technique is tested by a series of simulation tests. Test results show that the developed sBoost technique can improve the classification accuracy and prevent the overfitting problem effectively. [ABSTRACT FROM AUTHOR]

The backpropagation (BP) algorithm is the most commonly utilized training strategy for a feed-forward artificial neural network (FFANN). The BP algorithm, however, always leads to the problems of low convergence rate, high energy and poor generalization capability of FFANN. In this paper, motivated by the sparsity property of human neuron’ responses, we introduce a new sparse-response BP (SRBP) to improve the capacity of a FFANN by enforcing sparsity to its hidden units through imposing a supplemental L 1 term on them. The FFANN model learned from our algorithm is closely related to the real human and thus its mechanism fully complies with the human nervous system, i.e., sparse representation and architectural depth. Experiments on several datasets demonstrate that SRBP yields good performances on convergence rate, energy saving and generalization capability. [ABSTRACT FROM AUTHOR]

Automatic image annotation (AIA) is a task of assigning one or more semantic concepts to a given image and a promising way to achieve more effective image retrieval and analysis. It is a typical classification problem. Due to the semantic gap between low-level visual features and high-level image semantic, the performances of many existing image annotation algorithms are not satisfactory. This paper presents a novel AIA scheme based on improved quantum particle swarm optimization (IQPSO) algorithm for visual features selection (VFS) and an ensemble stratagem based on boosting technique to improve performance of image annotation approach. To maintain the population diversity, the measure method of population diversity and improvement operation are proposed. To achieve better performance of AIA scheme, the measure of population diversity is as a control condition of VFS process. The classification result of an ensemble classifier is as the final annotation result rather than individual classifier. The experimental results confirm that the proposed AIA scheme is very effectiveness. When using proposed AIA scheme over three image datasets respectively, the annotation results are satisfactory. [ABSTRACT FROM AUTHOR]

One of the primary objectives of Wireless Sensor Network (WSN) is to provide full coverage of a sensing field as long as possible. The deployment strategy of sensor nodes in the sensor field is the most critical factor related to the network coverage. However, the traditional deployment methods can cause coverage holes in the sensing field. Therefore, this paper proposes a new deployment method based on Multi-objective Immune Algorithm (MIA) and binary sensing model to alleviate these coverage holes. MIA is adopted here to maximize the coverage area of WSN by rearranging the mobile sensors based on limiting their mobility within their communication range to preserve the connectivity among them. The performance of the proposed algorithm is compared with the previous algorithms using Matlab simulation for different network environments with and without obstacles. Simulation results show that the proposed algorithm improves the coverage area and the mobility cost of WSN. [ABSTRACT FROM AUTHOR]

Non-convex regularization has attracted much attention in the fields of machine learning, since it is unbiased and improves the performance on many applications compared with the convex counterparts. The optimization is important but difficult for non-convex regularization. In this paper, we propose the Damping Proximal Coordinate Descent (DPCD) algorithms that address the optimization issues of a general family of non-convex regularized problems. DPCD is guaranteed to be globally convergent. The computational complexity of obtaining an approximately stationary solution with a desired precision is only linear to the data size. Our experiments on many machine learning benchmark datasets also show that DPCD has a fast convergence rate and it reduces the time of training models without significant loss of prediction accuracy. [ABSTRACT FROM AUTHOR]