Search

To compact the architecture of extreme learning machine (ELM), two incremental learning algorithms are proposed in this paper. The previous incremental learning algorithms for ELM recruit hidden nodes randomly, which is equivalent to implementing a random selection from a candidate set of infinite size. Hence, it is impossible to recruit good hidden nodes, and thus it usually requires more hidden nodes than traditional neural networks to achieve matched performance. To improve the quality of the hidden nodes recruited, an incremental learning algorithm for ELM is presented based on Gram--Schmidt process (GSI-ELM), which recruits the best hidden node from a random subset of fixed size via defining an evaluating criterion at each learning step. However, the “nesting effect” exists in the GSI-ELM, that is to say, the hidden nodes once recruited by GSI-ELM can not be later discarded. To treat this “nesting problem”, the improved GSI-ELM (IGSI-ELM) is generated with an elimination mechanism. At each learning step IGSI-ELM eliminates the worst hidden node from the already-recruited group if it is not the newly-recruited one. Finally, to verify the efficacy and feasibility of the proposed algorithms, i.e. GSI-ELM and IGSI-ELM, in this paper, experiments on regression and classification benchmark data sets are investigated.

Multi-task learning employs a shared representation of knowledge for learning several instances of the same problem. Multi-step time series problem is one of the most challenging problems for machine learning methods. The performance of a prediction model face challenges for higher prediction horizons due to the accumulation of errors. Cooperative coevolution employs in a divide and conquer approach for training neural networks and has been very promising for single step ahead time series prediction. Recently, co-evolutionary multi-task learning has been proposed for dynamic time series prediction. In this paper, we adapt co-evolutionary multi-task learning for multi-step prediction where predictive recurrence is developed to feature knowledge from previous states for future prediction horizon. The goal of the paper is to present a network architecture with predictive recurrence which is capable of multi-step prediction through a form of multi-task learning. We employ cooperative neuro-evolution and an evolutionary algorithm as baselines for comparison. The results show that the proposed method provides the best generalization performance in most cases. Comparison of results with the literature has shown to be promising which motivates further application of the approach for related real-world problems.

The voice conversion system modifies the speaker specific characteristics of the source speaker to that of the target speaker, so it perceives like target speaker. The speaker specific characteristics of the speech signal are reflected at different levels such as the shape of the vocal tract, shape of the glottal excitation and long term prosody. The shape of the vocal tract is represented by Line Spectral Frequency (LSF) and the shape of glottal excitation by Linear Predictive (LP) residuals. In this paper, the fourth level wavelet packet transform is applied to LP-residual to generate the sixteen sub-bands. This approach not only reduces the computational complexity but also presents a genuine transformation model over state of the art statistical prediction methods. In voice conversion, the alignment is an essential process which aligns the features of the source and target speakers. In this paper, the Mel Frequency Cepstrum Coefficients (MFCC) based warping path is proposed to align the LSF and LP-residual sub-bands using proposed constant source and constant target alignment. The conventional alignment technique is compared with two proposed approaches namely, constant source and constant target. Analysis shows that, constant source alignment using MFCC warping path performs slightly better than the constant target alignment and the state-of-the-art alignment approach. Generalized mapping models are developed for each sub-band using Radial Basis Function neural network (RBF) and are compared with Gaussian Mixture mapping model (GMM) and residual selection approach. Various subjective and objective evaluation measures indicate significant performance of RBF based residual mapping approach over the state-of-the-art approaches. HighlightsThe LSF fails to represent formant valleys but good for formant peaks. Hence, calculated warping path is not satisfactory to yield a better alignment.This LSF based warping overcome through a new alignment using MFCC based warping path, which improves the conversion performance of proposed system.Further, the existing techniques for mapping the LP-residual suffer from issues of artifacts generated in consecutive frames. The residual signal is also quite complex to map.In order to solve the high dimensionality issue of residual signal is reduced and the complexity of the model is decreased, the WPT and RBF pairs are employed.The experimental results prove that the proposed MFCC based warping path and the WPT-RBF based transformation for residual signal outperforms the state of the art methods of residual selection and GMM model.

Population coding has been established as the key mechanism for decoding sensory information received from periphery organs such as retinas and cochlear. In this paper a novel architecture is presented to embed population coding in neuromorphic hardware. A resonance based mechanism between two layers of neuron is utilized in the presented work. The mechanism discussed in this paper facilitates selective triggering of higher layer neurons which serves as target ensemble for evaluation of a particular input of interest. It has been shown that presented model can be used to detect any physical quantity (light intensity, temperature, etc.) feed to the sensor on the basis of population coding.

The paper presents a subspace neuro-fuzzy system with ordering of data items.The system assigns fuzzy weights to attributes in each fuzzy rule.The ordering technique makes the system robust to outliers.The presented system can outperform a subspace neuro-fuzzy system for noisy data. Neuro-fuzzy systems are known for their ability to both approximate and generalize presented data. In real life data sets not always all attributes (dimensions) of data are relevant or have the same importance. Some of them may be noninformative or unnecessary. This is why subspace technique is applied. Unfortunately this technique is vulnerable to noise and outliers that are often present in real life data. The paper describes a subspace neuro-fuzzy system with data ordering technique. Data items are ordered and assigned with typicalities. Data items with low typicalities have lower influence on the elaborated fuzzy model. This technique makes fuzzy models more robust to noise and outliers. The paper is accompanied by numerical experiments on real life data sets.

Recently, sparse representation has been applied in object tracking successfully. However, the existing sparse representation captures either the holistic features of the target or the local features of the target. In this paper, we propose a dual-scale structural local sparse appearance (DSLSA) model based on overlapped patches, which can capture the quasi-holistic features and the local features of the target simultaneously. This paper first proposes two-scales structural local sparse appearance models based on overlapped patches. The larger-scale model is used to capture the structural quasi-holistic feature of the target, and the smaller-scale model is used to capture the structural local features of the target. Then, we propose a new mechanism to associate these two scale models as a new dual-scale appearance model. Both qualitative and quantitative analyses on challenging benchmark image sequences indicate that the tracker with our DSLSA model performs favorably against several state-of-the-art trackers.

How to extract proper features is very important for synthetic aperture radar (SAR) target configuration recognition. However, most of feature extraction methods are hand-designed and usually can not achieve a satisfactory performance. In this paper, we propose a novel method based on the biologically inspired model to extract features automatically from limited data. Specifically, we learn episodic features (containing the key components and their spatial relations) and semantic features (i.e., semantic descriptions of the key components) which are two important types of features for the human cognition process. Episode features are learned through a deep neural network (DNN) and then semantic geometric features of the key components are defined. Moreover, SAR images are very sensitive to aspect angles. Therefore, we use episode features to estimate aspect angles of testing samples for the final recognition. This paper is a preliminary study and the preliminary experimental results on the moving and stationary target automatic recognition (MSTAR) database demonstrate the effectiveness of the proposed method.

[EN] This paper presents an approach of a social emotional model, which allows to extract the social emotion of a group of intelligent entities. The emotional model PAD allows to represent the emotion of an intelligent entity in 3-D space, allowing the representation of different emotional states. The social emotional model presented in this paper uses individual emotions of each one of the entities, which are represented in the emotional space PAD. Using a social emotional model within intelligent entities allows the creation of more real simulations, in which emotional states can influence decision-making. The result of this social emotional mode is represented by a series of examples, which are intended to represent a number of situations in which the emotions of each individual modify the emotion of the group. Moreover, the paper introduces an example which employs the proposed model in order to learn and predict future actions trying to influence in the social emotion of a group of people., This work is partially supported by the MINECO/FEDER TIN2015-65515-C4-1-R and the FPI grant AP2013-01276 awarded to Jaime-Andres Rincon.

With the advance of multimedia and pattern recognition based medical technologies, smart medical applications in smart hospital and smart medical for individual, such as disease diagnosis and health monitoring, play an important role in our life. However, the communication infrastructure of supporting these applications is a challenge. This paper studies the energy-efficient multicast routing problem in multi-hop wireless networks for these medical applications. Energy consumption has become the main problem of sustainable development of communication networks, particularly for these applications. How to carry out high energy-efficient communication is an important research topic for wired and wireless networks to implement green communications. This paper proposes an energy-efficient multicast routing approach to multi-hop wireless networks for smart medical applications. Different from previous methods, we aim at maximizing energy efficiency of networks. To this end, we make use of topology control and sleeping mechanism to obtain the optimal routing strategy with maximum network energy efficiency to construct the network multicast route. Simulation results show that the proposed approach is effective and feasible.

This paper addresses the problem of pose estimation in video sequences in which human pose changes drastically over time. Popular strategies for video pose estimation first yield multiple pose candidates for each frame and then achieve consistent pose estimation by enforcing temporal constraints across frames. To enrich pose candidates, previous methods typically employ local motion cues to propagate pose detections to adjacent frames. Reasonable pose proposals can be achieved only when the local motion estimation is accurate and good detections exist among adjacent frames, both of which are hard to be satisfied under drastic human pose changes. In this paper, we propose to propagate pose detections to entire video sequence through global motion cues which provide a long term holistic non-rigid motion transformation for the given video. We exploit the temporal continuity of both single parts and part pairs in the inference over a spatio-temporal model to stitch the reasonable trajectory fragments for each part and obtain the final pose estimation. Experimental results demonstrate remarkable performance improvement in comparison with the state-of-the-art methods.

Dictionary learning has played an important role in the success of sparse representation. Although discriminative synthesis dictionary learning for sparse representation with a high-computational-complexity l0 or l1 norm constraint has been well studied for image classification, jointly and discriminatively learning an analysis dictionary and a synthesis dictionary is still in its infant stage. As a dual of synthesis dictionary, the recently developed analysis dictionary can provide a complementary view of data representation, which can have a much lower time complexity than sparse synthesis representation. Although several class-specific analysis-synthesis dictionary, which may have a big correlation between different classes' dictionaries, have been developed, how to learn a more compact and discriminative universal analysis-synthesis dictionary is still open. In this paper, to provide a more complete view of discriminative data representation, we propose a novel model of discriminative analysis-synthesis dictionary learning (DASDL), in which a linear classifier based on the coding coefficient is jointly learned with the dictionary pair, thus the performance of the classifier and the representational power of the dictionary pair being considered at the same time by the same optimization procedure. The size of the learned dictionaries can be very small since the analysis-synthesis dictionary is shared by all class data. An iterative algorithm to efficiently solve the proposed DASDL is presented in this paper. The experiments on face recognition, gender classification, action recognition and image classification clearly show the superiority of the proposed DASDL.

With the advent of multimedia and pattern recognition based medical technology, intelligence medical applications in smart hospital, smart clinic, and smart medical for Individual, such as disease diagnosis, rehabilitation, health monitoring, and so on, have received extensive attentions. However, the communication infrastructure of supporting these applications is facing a larger challenge. This paper studies the channel allocation problem in cognitive wireless networks for these medical applications. Channel allocations are of importance for cognitive wireless networks. How to assign the appropriate channels to each cognitive user is fairly challenging. In this paper we study this problem by combining collaborations of network nodes. We take the maximum network connectivity as an optimal objective to build our collaborative channel allocation model. Our approach takes in account the channel idle status and the transmission radiuses of cognitive users. We allocate firstly the idle channel to bottleneck node and assign the suited channels to all other cognitive users as possible. In such a way, we present three channel allocation strategies with the maximum network connectivity via the omni-directional and direction antennas. We also propose the corresponding algorithm to perform the channel allocation for all cognitive users in cognitive wireless networks. Simulation results show that our approach is promising.

Remote medical resources configuration and management involves complex combinatorial Multi-Objective Optimization problem, whose computational complexity is a typical NP problem. Based on the MOEA/D framework, this paper applies the two-way local search strategy and the new selection strategy based on domination amount and proposes the IMOEA/D framework, following which each individual produces two individuals in mutation. In this paper, by using a new selection strategy, the parent individual is compared with two mutated offspring individuals, and the more excellent one is selected for the next generation of evolution. The proposed algorithm IMOEA/D is compared with eMOEA, MOEA/D and NSGA-II, and experimental results show that for most test functions, IMOEA/D proposed is superior to the other three algorithms in terms of convergence rate and distribution.

With the increasing public consciousness in environmental protection, the green partner selection problem (G-PSP) is an important issue in virtual enterprises. In this paper, the green criterion is introduced to partner selection problem (PSP), and a green partner selection model based on six criteria is proposed. As PSP has been proven to be an NP problem, and G-PSP cannot be solved in reasonable time by traditional methods. In this paper, an improved algorithm I-GSA/PSO that combines gravitational search algorithm and particle swarm optimization is developed to solve G-PSP in virtual enterprises. Experimental results show that I-GSA/PSO is effective and outperforms other evolutionary algorithms in solving G-PSP.

The person re-identification is an active research branch of the computer vision and attracts many researchers study on it. However, because of the variance in viewpoints, illumination, pedestrians' pose and some of other factors, building a robust person re-identification algorithm in open-world is rather challenging. In this paper, we propose two algorithms, indiscriminative patches trim strategy and multi-instance multi-label learning method, to fix the person re-identification in closed short-term surveillance network. There are two main contributions of this paper: first, we define the discriminative region in a person's image. Correspondingly, an adapted canopy-kmeans algorithm is proposed to evaluate the discriminability of image patches which form the discriminative regions. Besides, two strategies, in local image and in global gallery dataset, are proposed to filter out distractions. Finally, the true discriminative patches are employed to compute the similarity between two images, and then used to re-identify pedestrians. Second, we first introduce the multi-instance multi-label learning methods into re-identification, and hence we propose a framework of solving person re-identification by applying MIMLL. We employ two MIMLL methods, the MIMLBOOST and the MIMLSVM, to detect attributes in each image, and those detected attributes contribute to recognize different pedestrians. Experiments on two benchmark datasets show the competitive performance of the two proposed algorithms.

This paper presents a new feature extraction method, compass local binary pattern (CoLBP) for facial gender recognition. To achieve robustness, the proposed method first computes directional edge responses using eight Kirsch compass masks. Then, the spatial relationships among the neighboring pixels in each edge response are exploited independently with the help of local binary pattern (LBP) to enhance the discrimination capability. Finally, spatial histograms computed from these LBP images are concatenated to build a face descriptor. Our proposed descriptor efficiently extracts discriminating information from four different levels, including gradient, regional, global and directional level. The proposed method was evaluated on three datasets (color FERET, LFW, and Adience) containing facial photographs. In spite of a wide range of challenges (low resolution, variations in pose, expression, and illumination) present in the datasets, the proposed method provided promising classification performance in comparison with several existing benchmark methods, thereby validating its robustness. Moreover, this paper also investigates the gender recognition of facial sketches. The experiments carried out on two facial sketch datasets including CUFS and CUFSF also demonstrated better classification performance of the proposed method. A new face descriptor called compass LBP (CoLBP) is proposed.The proposed descriptor effectively combines the properties of Kirsch masks and LBP.Gender classification of sketch images is carried out on two sketch datasets.The CoLBP provides promising classification performance on two different modalities.The computational complexity of CoLBP is much lower than LGBPHS and MBP.

Kernel based extreme learning machine (K-ELM) has better generalization performance than basic ELM with less tuned parameters in most applications. However the original K-ELM is lack of sparseness, which makes the model scale grows linearly with sample size. This paper focuses on sparsity of K-ELM and proposes recursive reduced kernel based extreme learning machine (RR-KELM). The proposed algorithm chooses samples making more contribution to target function to constitute kernel dictionary meanwhile considering all the constraints generated by the whole training set. As a result it can simplify model structure and realize sparseness of K-ELM. Experimental results on benchmark datasets show that no matter for regression or classification problems, RR-KELM produces more compact model structure and higher real-time in comparison with other methods. The application of RR-KELM for aero-engine fault pattern recognition is also given in this paper. The simulation results demonstrate that RR-KELM has a high recognition rate on aero-engine fault pattern based on measurable parameters of aero-engine. A new algorithm is proposed in this paper to realize the sparsity of K-ELM.The algorithm makes better use of data information and builds more parsimonious models.The algorithm is used to recognize the aero-engine fault pattern and it works well.

In this paper, we tackle the problem of depth estimation from a single image, which is essential for understanding 3D scene structure and can promote the development of visual applications. Markov Random Field (MRF) related depth estimation methods have attracted extensive attentions due to their ability of building structural relationships on outputs. Almost all of them employed engineered low-level absolute and relative features for constructing MRF. However, the engineered features are not capable of producing accurate depths for various scenes. In this paper, we propose Convolutional Conditional Random Field Network (CCRFN) consisting of feature learning and depth tuning components. In feature learning component, we build two Convolutional Neural Network (CNN) architectures to learn absolute and relative features from raw images. In depth tuning component, the learned features are fed into Conditional Random Field (CRF) to generate the depths of all pixels in an image by optimizing a well-defined energy function. CCRFN has two advantages that (1) it does not need hand-crafted features and (2) it models the depths at individual points as well as the relationship between them in the deep model. Experiments on widely used Make3D dataset show that CCRFN outperforms state-of-the-art methods on three evaluation criteria and generates more accurate depth maps with sharper object boundaries.

Very recently, vehicle detection in satellite images has become an emerging research topic with various applications ranging from military to commercial systems. However, it retains as an open problem, mainly due to the complex variations in imaging conditions, object intra-class changes, as well as due to its low-resolution. Coming with the rapid advances in deep learning for feature representation, in this paper we investigate the possibility to exploit deep neural features towards robust vehicle detection. In addition, along with the rapid growth in the data volume, new classification methodology is also demanded to explicitly handle the intra-class variations. In this paper, we propose a vehicle detection framework, which combines Deep Convolutional Neural Network (DNN) based feature learning with Exemplar-SVMs (E-SVMS) based, robust instance classifier to achieve robust vehicle detection in satellite images. In particular, we adopt DNN to learn discriminative image features, which has a high learning capacity. In our practice, the leverage of DNN has achieve significant performance boost by comparing to a serial of handcraft designed features. In addition, we adopt E-SVMs based robust classifier to further improve the classification robustness, which can be considered as an instance-specific metric learning scheme. By conducting extensive experiments with comparisons to a serial of state-of-the-art and alternative works, we further show that the combination of both schemes can benefit from each other to jointly improve the detection accuracy and effectiveness.

As the growing popularity of the Internet, the online social network has become an indispensable part and played a more and more important role in our daily life. People in the online social networks would link with their friends through chatting, e-mail, posting, commenting directly, which are built on the mature relationships in reality. However, this link-ship provides only a coarse representation of relationship rather than revealing the relationship strengths. Accurate measurement of social relationship could be applied to many applications, such as project or paper review. In this paper, we propose a probabilistic graphical model to measure the relationship strengths between different users in the social network by taking consideration of the similarities among users profiles, co-occurrence of user names and interaction activities in different activity fields. HighlightsThe highlights of this paper is that we proposed a probabilistic graphical model to measure the relationship strength between different users in social network by taking consideration of three information sources: similarity of users' profiles, co-occurrence of user names and interaction activities on different activity fields.

Engineering Asset Management (EAM) is a recently attractive discipline and it aims to address valuable contributions of asset management to organization's success. As of today, there is no specific method to evaluate performance of EAM standards. This paper aims to fill this gap and rank performance of asset management automatically after conducting survey, instead of evaluating questionnaires, analyzing results and ranking performances with a tedious process. Hence, it is necessary to develop intelligent data classification to simplify the whole procedure. Among many supervised learning methods, support vector machine attracts much attention for binary classification problems and its extension, namely multiple support vector machines, is able to solve multiclass classification problems. It is crucial to find optimal parameters of support vector machines prior to their use for prediction of unknown testing data sets. In this paper, novel Bayesian inference on optimal parameters of support vector machines is proposed. Firstly, a state space model is constructed to find the relationship between parameters of support vector machines and guess cross-validation accuracy. Here, the guess cross-validation accuracy aims to prevent support vector machines from overfitting. Secondly, particle filter is introduced to iteratively find posterior probability density functions of the parameters of support vector machines. Then, optimal parameters of support vector machines can be found from the posterior probability density functions. Ultimately, survey data collected from industry are used to validate the effectiveness of the proposed Bayesian inference method. Comparisons with some randomly selected parameters are conducted to highlight the superiority of the proposed method. The results show that the proposed Bayesian inference method can result in both high training and testing accuracies.

Online social network presents a great opportunity to analyze user behavior and mine the implicit personality traits from the social network data. Considering the personality recognition as a multi-label classification problem, this paper proposes a new probabilistic topic model (PT-LDA model) to predict the personality traits within the framework of Five Factor Model. The proposed model extends the Latent Dirichlet Allocation (LDA) model to integrate the n-gram features into few latent topics and each topic is characterized by not only the multinomial distribution over words but also the Gaussian distributions over personality traits. This paper develops a Gibbs-EM algorithm to solve the proposed model iteratively based on Gibbs sampling and expectation maximization. Quantitative evaluation shows that PT-LDA is more accurate, efficient and robust than several baselines. Our experiment also shows that the proposed model can be used to extract the interpretable topics associated with each personality trait, which provides a new way to uncover user behaviors in online social network.

In today׳s society where audio–visual content is ubiquitous, violence detection in movies and Web videos has become a decisive functionality, e.g., for providing automated youth protection services. In this paper, we concentrate on two important aspects of video content analysis: Time efficiency and modeling of concepts (in this case, violence modeling). Traditional approaches to violent scene detection build on audio or visual features to model violence as a single concept in the feature space. Such modeling does not always provide a faithful representation of violence in terms of audio–visual features, as violence is not necessarily located compactly in the feature space. Consequently, in this paper, we target to close this gap. To this end, we present a solution which uses audio–visual features (MFCC-based audio and advanced motion features) and propose to model violence by means of multiple (sub)concepts. To cope with the heavy computations induced by the use of motion features, we perform a coarse-to-fine analysis, starting with a coarse-level analysis with time efficient audio features and pursuing with a fine-level analysis with advanced features when necessary. The results demonstrate the potential of the proposed approach on the standardized datasets of the latest editions of the MediaEval Affect in Multimedia: Violent Scenes Detection (VSD) task of 2014 and 2015.

People counting is one of the key techniques in video surveillance. This task usually encounters many challenges in crowded environment, such as heavy occlusion, low resolution, imaging viewpoint variability, etc. Motivated by the success of R-CNN (Girshick et al., 2014) 1 on object detection, in this paper we propose a head detection based people counting method combining the Adaboost algorithm and the CNN. Unlike the R-CNN which uses the general object proposals as the inputs of CNN, our method uses the cascade Adaboost algorithm to obtain the head region proposals for CNN, which can greatly reduce the following classification time. Resorting to the strong ability of feature learning of the CNN, it is used as a feature extractor in this paper, instead of as a classifier as its commonly-used strategy. The final classification is done by a linear SVM classifier trained on the features extracted using the CNN feature extractor. Finally, the prior knowledge can be applied to post-process the detection results to increase the precision of head detection and the people count is obtained by counting the head detection results. A real classroom surveillance dataset is used to evaluate the proposed method and experimental results show that this method has good performance and outperforms the baseline methods, including deformable part model and cascade Adaboost methods.

In this paper, we propose a transfer subspace learning approach cross-dataset facial expression recognition. To our best knowledge, this problem has been seldom addressed in the literature. While many facial expression recognition methods have been proposed in recent years, most of them assume that face images in the training and testing sets are collected under the same conditions so that they are independently and identically distributed. In many real applications, this assumption does not hold as the testing data are usually collected online and are generally more uncontrollable than the training data. Hence, the testing samples are likely different from the training samples. In this paper, we define this problem as cross-dataset facial expression recognition as the training and testing data are considered to be collected from different datasets due to different acquisition conditions. To address this, we propose a transfer subspace learning approach to learn a feature subspace which transfers the knowledge gained from the source domain (training samples) to the target domain (testing samples) to improve the recognition performance. To better exploit more complementary information for multiple feature representations of face images, we develop a multi-view transfer subspace learning approach where multiple different yet related subspaces are learned to transfer information from the source domain to the target domain. Experimental results are presented to demonstrate the efficacy of these proposed methods for the cross-dataset facial expression recognition task.

Internet of Things (IoT), an emerging computing paradigm which interconnects various ubiquitous things is facilitating the advancement of computational intelligence. This paper aims at investigating the computation intelligence extraction approach with focus on the dynamic k-clique mining that is an important issue in social network analysis. The k-clique detection problem as one of the fundamental problems in computer science, can assist us to understand the organization style and behavioral patterns of users in social networks. However, real social networks usually evolve over time and it remains a challenge to efficiently detect the k-cliques from dynamic social networks. To address this challenge, this paper proposes an efficient k-clique dynamic detection theorem based on triadic formal concept analysis (TFCA) with completed mathematical proof. With this proposed detection theorem, we prove that the k-cliques detection problem is equivalent to finding the explicit k-cliques generated from k-triadic equiconcepts plus the implicit k-cliques derived from its high-order triadic equiconcepts. Theoretical analysis and experimental results illustrate that the proposed detection algorithm is efficient for finding the k-cliques and exploring the dynamic characteristics of the sub-structures in social networks.

With the explosive growth of user-generated multimedia resources in the big data era (e.g., video, audio, text or even their combinations), bridging the semantic gap between low-level features and high-level semantics in multiple modes of data is a critical and indispensable issue. In this paper, we exploit a popular type of metadata called collaborative tags to address this issue. Specifically, we identify basic level concepts and then construct ontologies from the collaborative tags. The generated ontologies can be employed to organize and index user-generated multimedia data. Existing research lacks a principle to supervise the ontology extraction from a human perspective. In contrast, we borrow the idea of a family of concepts called basic level concepts from cognitive psychology. These basic level concepts are frequently used by people in their daily life and when organizing human knowledge. In this paper, we extract ontologies with basic level concepts from collaborative tags. Furthermore, we model the effect of context and present a method for context-aware basic level concepts detection for ontology learning. To the best of our knowledge, this is the first work on context-aware basic level concept detection in collaborative tagging for the construction of ontologies. To evaluate the proposed method, experiments were conducted on real datasets using the Open Directory Project (ODP) as a benchmark. The experimental results illustrate that ontologies extracted by our approach are more rational and human-oriented.

Recent years have witnessed the fast development of multimedia platforms in China, such as Youku, LeTV and Weibo. Images and videos are usually uploaded with textual descriptions, such as titles and introductions of these media. These texts are the key to multimedia content understanding, and this paper is dedicated to multimedia understanding with visual content entailment via recognizing semantic entailment in these texts. In fact, the natural language processing community has been manifesting increasing interest in semantic entailment recognition in English texts. Yet, so far not much attention has been paid to semantic entailment recognition in Chinese texts. Therefore, this paper investigates on multimedia semantic entailment with Chinese texts. Recognizing semantic entailment in Chinese texts can be cast as a classification problem. In this paper, a classification model is constructed based on support vector machine to detect high-level semantic entailment relations in Chinese text pair, including entailment and non-entailment for the Binary-Class and forward entailment, reverse entailment, bidirectional entailment, contradiction and independence for the Multi-Class. We explore different semantic feature combinations based on three kinds of Chinese textual features, including Chinese surface textual, Chinese lexical semantic and Chinese syntactic features, and utilize them to feed our classification model. The experiment results show that the accuracy of our classification model for semantic entailment recognition with the feature combination using all the three kinds of Chinese textual features achieves a much better performance than the baseline in Multi-Class and slightly better results than the baseline in the Binary-Class.

In the traditional Support Vector Machine Plus (SVM+), the grouping method has great randomness and only takes into account part of the structural information of the dataset. In order to overcome these shortcomings, in this paper, we propose a novel framework termed as FCSVM+ to improve the performance of SVM+ by combining clustering technique and feature selection. The new framework strategy is expected to not only take fully into account the structural information of the training data, but also partition the training data into more meaningful groups. To prove the advantage of the framework, in particular, we adopt two simplest feature selection methods, i.e. F-score and Laplacian score methods, to select the features, then apply a recently proposed clustering technique to get a better partition of training data by the selected features, in which the number of clusters could be found automatically. Three major contributions of this paper can be concluded as: (1) improving the performance of the existing SVM+ classifier; (2) extracting the potential structural information of the training data by using more feature attributes instead of one; (3) replacing the truncation method in SVM+ with a clustering technique. The comprehensive experimental results on the UCI benchmark datasets illustrate the validity and advantage of our approach.

Yarbus' pioneering work in eye tracking has been influential to methodology and in demonstrating the apparent importance of task in eliciting different fixation patterns. There has been renewed interest in Yarbus' assertions on the importance of task in recent years, driven in part by a greater capability to apply quantitative methods to fixation data analysis. A number of recent research efforts have examined the extent to which an observer's task may be predicted from recorded fixation data. This body of recent work has raised a number of interesting questions, with some investigations calling for closer examination of the validity of Yarbus' claims, and subsequent efforts revealing some of the nuances involved in carrying out this type of analysis including both methodological, and data related considerations. In this paper, we present an overview of prior efforts in task prediction, and assess different types of statistics drawn from fixation data, or images in their ability to predict task from gaze. We also examine the extent to which relatively general task definitions (free-viewing, object-search, saliency-viewing, explicit saliency) may be predicted by spatial positioning of fixations, features co-located with fixation points, fixation dynamics and scene structure. This is accomplished in considering the data of Koehler et al. (2014) 30 affording a larger scale, and qualitatively different corpus of data for task prediction relative to existing efforts. Based on this analysis, we demonstrate that both spatial position, as well as local features are of value in distinguishing general task categories. The methods proposed provide a general framework for highlighting features that distinguish behavioural differences observed across visual tasks, and we relate new task prediction results in this paper to the body of prior work in this domain. Finally, we also comment on the value of task prediction and classification models in general in understanding facets of gaze behaviour.

High resolution remote sensing image captured by the satellites or the aircraft is of great help for military and civilian applications. In recent years, with an increasing amount of high resolution remote sensing images, it becomes more and more urgent to find a way to retrieve them. In this case, a few methods based on the statistical information of the local features are proposed, which have achieved good performances. However, most of the methods do not take the topological structure of the features into account. In this paper, we propose a new method to represent these images, by taking the structural information into consideration. The main contributions of this paper include: (1) mapping the features into a manifold space by a Lipschitz smooth function to enhance the representation ability of the features; (2) training an anchor set with several regularization constrains to get the intrinsic manifold structure. In the experiments, the method is applied to two challenging remote sensing image datasets: UC Merced land use dataset and Sydney dataset. Compared to the state-of-the-art approaches, the proposed method can achieve a more robust and commendable performance.

A social network is a social structure made up of a set of social actors (such as individuals or organizations) and a set of the dyadic ties between these actors. By contrast, for the fixed time duration the size of digital video would be much bigger than that of digital sound. Consequently, providers of social network services can offer real-time chatting among users which could offer satisfactory experiences for users. As one of the most popular content-based social network services (SNS), chatting service plays an important role in current big data era. Also average data packets' transmission via networks is another significant traffic. So how to offer satisfactory Quality of Service (QoS) for users is the key problem which will be solved for SNS provider. For real time communication among users, end-to-end time delay seems to be critical in user's experience. Therefore modeling and evaluating social network systems is an important and urgent issue which offers quantitative basis of SNS with high quality for users. For social network system, the scalability and robust are important for both service provider and users under the circumstance of a large number of users. On the basis of performance evaluation of social network system of one user case, we construct the SPN model and conduct numerical analysis to discover and report the performance with the addition of users. By taking hybrid traffic containing voice and data into account, this paper constructed a Stochastic Petri Net (SPN) model for data and ON/OFF voice traffic for social network system. Then, average time delay of the system was analyzed and model-based simulation is conducted with Stochastic Petri Net Package (SPNP) 6.0. Furthermore, for different parameters of burst rate, idle rate, number of data packets, traffic load and buffer size, variation trends on average time delay are derived thereby. On the basis of the work in this paper, further research on heterogeneous objects of social network systems can be carried on.

This paper is concerned with the problem of µ-stability analysis of memristor-based recurrent neural networks with the effects of bounded and unbounded leakage time-varying delays. A new idea of µ-stability analysis of neural networks is given first, then by means of the linear matrix inequality (LMI) approach, stability criteria are presented. Two types of memductance functions is used to derive the proposed stability results. Obviously, the memristive neural network with different memductance functions is a state-dependent switched system or a state-dependent continuous system, which is the generalization of those for conventional artificial neural networks. Under the framework of Filippov solutions, µ-stability analysis of solutions for functional differential inclusions and memristor-based neural networks can be guaranteed by constructing suitable Lyapunov-Krasovskii functionals, suitable inequalities and LMIs. The dynamic analysis in this paper utilizes the theory of set-valued maps and functional differential equations with discontinuous right-hand sides of Filippov type. Leakage time-varying delay is considered in this paper to be bounded, unbounded and differentiable. Taking into account of the information of the neuron activation functions and unbounded time-varying delays, several improved results have been obtained in terms of LMIs and these are tested by MATLAB LMI toolbox. The model based on the memristor widens the relevance scope for the design of neural networks, and the new effective results also enrich the toolbox for the qualitative analysis of neural networks. Finally, numerical examples are provided to demonstrate the effectiveness of the proposed theoretical results.

In this paper, an approach to learn a robust sparse representation dictionary for face recognition is proposed. As well known, sparse representation algorithm can effectively tackle slight occlusion problems for face recognition. However, if images are corrupted by heavy noise, performance will be not guaranteed. In this paper, to enhance the robustness of sparse representation to serious noise in face images, we integrate low rank representation into dictionary learning to alleviate the influence of unfavorable factors such as large scale noise and occlusion. Among which we extract eigenfaces by singular value decomposition (SVD) from the low rank pictures to reduce dictionary atoms and, thereby, optimize the efficiency of improved algorithm. Otherwise, we characterize each image using the histogram of orientated gradient (HOG) feature which has been proven to be an effective descriptor for face recognition in particular. The performance of the proposed Low-rank and HOG feature based ESRC (LH_ESRC) algorithm on several popular face databases such as the Extended Yale B database and CMU_PIE face database shows the effectiveness of our method. In addition, we evaluate the robustness of our method by adding different proportions of randomly noise and block occlusion and real disgusts. Experimental results illustrate the benefits of our approach.

In this paper, we focus on unsupervised feature selection. As we have known, the combination of several feature units into a whole feature vector is broadly adopted for effective object representation, which may inevitably includes some irrelevant/redundant feature units or feature dimensions. Most of the traditional feature selection models can only select the feature dimensions without concerning the intrinsic relationship among different feature units. By taking into consideration the group sparsity of feature dimensions and feature units based on an l 2 , 1 minimization, we propose a new unsupervised feature selection model, unsupervised deep sparse feature selection (UDSFS) in this paper. In comparison with the state-of-the-arts, our UDSFS model can not only select the most discriminative feature units but also assign proper weight to the useful feature dimensions concurrently; moreover, the efficiency and robustness of our UDSFS can be also improved without extracting the discarded irrelevant feature units. For model optimization, we introduce an efficient iterative algorithm to solve the non-smooth, convex model and obtain a global optimization with the convergence rate as O ( 1 / K 2 ) (K is the iteration number). For the experiments, a new medical endoscopic image dataset, Abnormal Endoscopic Image Detection dataset (AEID), is built for evaluation; we also test our model using two public UCI datasets. Various experiments and comparisons with other state-of-the-arts justified the effectiveness and efficiency of our UDSFS model.

In this paper we develop a set of parallel algorithms for image segmentation basing on the authors' former work, the self-organizing map (SOM) based vector quantization (VQ) approach, by extending the method from serial computation to parallel processing, in order to accelerate the computation process. The parallel segmentation scheme is composed of a group of parallel algorithms for implementing the whole segmentation process, including parallel classification of the image into edge and non-edge pattern vectors, parallel training of an SOM network, parallel segmentation of the image by using the trained SOM model with VQ method, and adaptive parallel estimation of the segment number of the image being processed. In the paper, all the parallel algorithms have been implemented on graphic processing units (GPU) and applied to segmenting the human brain MRI images. The experimental results obtained in the work show that, compared with the original serial method implemented on CPU, the proposed parallel approach can achieve a significant improvement on the computation efficiency with overall speedup ratios increasing from 28.81 to 89.12 as image sizes increasing from 128×128 to 1024×1024, while keeping the segmentation performance unchanged.

For remote sensing images, whose spectral signatures are intricate, the traditional edge detection methods cannot obtain satisfactory results. This paper takes the space computing capacity of Cellular Automata (CA) and the data pattern search ability of Extreme Learning Machine (ELM) into account and puts forward an Extreme Learning Machine based on Cellular Automata (ELM-CA) of edge detection for remote sensing images. This model can extract evolution rules of Cellular Automata. On the basis of the rules, false edges are removed and purer edge map is obtained. The result of the simulation experiment shows that the performance of the method suggested by this paper is much better compared to other edge detection arithmetic operators. It can prove that ELM-CA is an ideal method of remote sensing image edge detection.

Data fitting and prediction are important in many domains. As computing power improves, artificial intelligence fitting methods such as the WASD (weights and structure-determination) neuronet become more operable. Although the WASD neuronet has been applied to several other issues, its application on fitting data needs to be explored and discussed more specifically. This paper is committed to introducing the WASD-neuronet model activated by Chebyshev polynomials of class 1 for data fitting and to exploring its capability of data prediction. The learning-checking method and the concept of global minimum point are introduced to improve the prediction performance and extend the application of the WASD-neuronet model. Applying such a model to Asian population prediction substantiates its excellent performance. With numerical experiments validating the predicting performance and a final prediction based on historical data, this paper presents a reasonable tendency of Asian population.

It is well known that the Leave-One-Out Cross-Validation (LOO-CV) is a highly reliable procedure in terms of model selection. Unfortunately, it is an extremely tedious method and has rarely been deployed in practical applications. In this paper, a highly efficient Leave-One-Out Cross-Validation (LOO-CV) formula has been developed and integrated with the popular Regularized Extreme Learning Machine (RELM). The main contribution of this paper is the proposed algorithm, termed as Efficient LOO-CV-based RELM (ELOO-RELM), that can effectively and efficiently update the LOO-CV error with every regularization parameter and automatically select the optimal model with limited user intervention. Rigorous analysis of computational complexity shows that the ELOO-RELM, including the tuning process, can achieve similar efficiency as the original RELM with pre-defined parameter, in which both scale linearly with the size of the training data. An early termination criterion is also introduced to further speed up the learning process. Experimentation studies on benchmark datasets show that the ELOO-RELM can achieve comparable generalization performance as the Support Vector Machines (SVM) with significantly higher learning efficiency. More importantly, comparing to the trial and error tuning procedure employed by the original RELM, the ELOO-RELM can provide more reliable results by the virtue of incorporating the LOO-CV procedure.

This paper proposes a self-representation method for kNN (k nearest neighbors) classification. Specifically, this paper first designs a self-reconstruction method to reconstruct each data point by all the data, and the derived reconstruction coefficient is then used for calculating the k value for each training sample. Furthermore, a decision tree is built with the resulting k values for each data point to output labels of the training samples. With the built decision tree, the proposed method classifies test samples. Finally, the experimental results on real datasets showed the proposed method outperformed the state-of-the-art methods.

Stereo matching is one of the most important and fundamental topics in computer vision. Encouraging self-similar pixels to be assigned to the same label has been proved to be effective for stereo. A typical way of taking advantage of self-similarity is performing a color segmentation on the image and motivating the pixels within each segment to share an identical label. However, some cases cannot be handled by image segmentation, such as the pixels in disconnected regions. This paper proposes a stereo method based on the assumption, that a 3D scene is a collection of a few smooth surfaces and a few classes of reflective materials, such that the 3D points belonging to an identical material are likely to lie on a small number of surfaces and the 3D points lying on a single surface belong to a few classes of reflective materials. Each material is expected to have specific albedo properties. This paper presents two methods for classifying the albedo properties depending on whether the illumination environment is known, without recovering the albedo parameters. The proposed model is formulated as an energy function incorporating some new priors, that is optimized via fusion move algorithm.

In this paper a people recognition algorithm is proposed, in which the color processing mechanism is inspired by the biological visual system. The algorithm is constructed in two parts. In the first part a spiking neural network is proposed to extract the color features of the people images which are captured from videos, and a set of new features is generated by fusing the color features and color moments. In the second part, after a feature reduction, a Support Vector Machine is trained and then used to recognize a specific people. The algorithm has been successfully applied to recognize people in CASIA Database with a high recognition rate. In order to evaluate performance and analyze characteristics of people recognition algorithms in multi-camera scenes, Multi-Camera Video (MCV) dataset is made in this paper. It is used to evaluate and analyze the proposed method and a set of characteristics of the proposed algorithms are obtained. Experimental results demonstrate that the algorithm is comparable to state-of-the-art approaches in terms of accuracy and the direction for further improvement of the proposed algorithm is provided.

We propose a robust object tracking algorithm based on local region sparse appearance model in this paper. In this algorithm, the object is divided into several sub-regions, and the sparse dictionaries are obtained by clustering in each sub-region. Therefore spatial structure information of the object can be captured well, and the change of object appearance can be also resisted effectively. First, the object is divided into many small patches. Then the object is divided into several sub-regions according to patch distribution again. The establishment of object dictionary base is based on combination of the dictionaries from all the sub-regions, and then space alignment between different parts of the object can be achieved. Meanwhile, noise removal and other operations in the existing sparse reconstruction error maps are performed to retain valuable information. In the updating framework, a novel flexible template set update mechanism is introduced in this paper. In this update mechanism, valuable object samples will be put into the template set. If samples are not valuable, they should not be put into the template set, even when the template set is not full. Then we use patch sparse coefficient histogram of updated templates to extract time domain information of the object in the form of weighted sum. Therefore, it can provide a reliable template basis for obtaining good candidate object. In addition, when tracking result deviates from the actual position of the object, we use a dynamic sub-region resampling method based on cosine angle to correct the position deviation timely. Therefore this method can effectively prevent the object from being completely lost. Both qualitative and quantitative evaluations on challenging video sequences demonstrate that the proposed tracking algorithm performs favorably against several state-of-the-art methods.

With the wide use of consumer electronics and the rapid development of online shopping, more and more ad videos are developed for IDTV and mobile users. However, a huge amount of time spending on the Internet advertising somehow brings users uncomfortable viewing experience rather than effectively generates high consumption of advertised products. Therefore, it is urgent to find a viewer-friendly and advertiser-beneficial solution to launch ads. This paper is the first attempt to improve the effectiveness of advertising through combining online shopping information with an ad video and directing viewers to proper online shopping places. The proposed ActiveAd framework includes four main components. Firstly, an ad video analysis component detects both syntactic and semantic elements from ad videos, e.g. FMPIs (Frame Marked with Production Information), visual concepts, and textual keywords within the ad videos. Our ad video analysis provides a comprehensive solution to extract meaningful elements from ad videos. Secondly, a visual linking by search component is proposed to collect websites which contain similar images as FMPIs. Features used for the visual search are weighted and fused in order to ensure the uniformity of search results. Thirdly, different kinds of tags and product categories extracted from collected websites are aggregated in order to identify the representative text of the product. Finally, query keywords are selected through calculating cosine similarity from two kinds of keywords, i.e. keywords identified from tag aggregation and keywords obtained through ad video analysis (visual concept detection and textual keyword detection). A query vector is generated by selected keywords and used to retrieve product online. In this paper, a powerful cross-media contextual search including visual search, tag aggregation and textual search is achieved with the help of ad video analysis. Experiments demonstrate that our proposed ActiveAd achieves product recommendation effectively and efficiently.

Auto-encoder-a tricky three-layered neural network, known as auto-association before, constructs the "building block" of deep learning, which has been demonstrated to achieve good performance in various domains. In this paper, we try to investigate the dimensionality reduction ability of auto-encoder, and see if it has some kind of good property that might accumulate when being stacked and thus contribute to the success of deep learning.Based on the above idea, this paper starts from auto-encoder and focuses on its ability to reduce the dimensionality, trying to understand the difference between auto-encoder and state-of-the-art dimensionality reduction methods. Experiments are conducted both on the synthesized data for an intuitive understanding of the method, mainly on two and three-dimensional spaces for better visualization, and on some real datasets, including MNIST and Olivetti face datasets. The results show that auto-encoder can indeed learn something different from other methods. Besides, we preliminarily investigate the influence of the number of hidden layer nodes on the performance of auto-encoder and its possible relation with the intrinsic dimensionality of input data.

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

This paper delivers a preliminary attempt towards person-independent appearance-based gaze estimation. Conventional methods need to assume training and test data collected from the same person, otherwise eye shape difference due to individuality will affect the estimation severely. To solve this problem, the key idea in this paper is to extract from eye images more advanced eye features, which helps learn a person-independent relationship between eye gaze change and eye appearance variation. To this end, we propose employing the advantages of recent sparse auto-encoding techniques. We partition any eye image into small patches which can overlap with each other. With patches from many images, we learn a codebook comprising a set of bases, which can reconstruct any eye image patch with sparse coefficients. By examining these coefficients, we can analyze the eye shape more effectively. Finally, we produce the eye features by pooling the coefficients at different scales, and then combine these subfeatures from different codebooks. Experimental results show that the proposed method achieves good accuracy on a public dataset and it also outperforms conventional methods by a large margin. HighlightsThis submission proposes a novel gaze estimation method from eye images. The highlights include:Computation of patch-based sparse representation of eye appearance.Final eye feature by using multiple codebooks and spatial pyramid pooling.Modeling and solving the problem by using the proposed eye feature.The method outperforms previous ones in person-independent gaze estimation.

This paper proposes ELMOEA/D, a surrogate-assisted MOEA, for solving costly multi-objective problems in small evaluation budgets. The proposed approach encompasses a state-of-the-art MOEA based on decomposition and Differential Evolution (MOEA/D-DE) assisted by Extreme Learning Machines (ELMs). ELMOEA/D is tested in instances from three well-known benchmarks (ZDT, DTLZ and WFG) with 5-60 decision variables, 2 and 5 objectives. The ELMOEA/D's performance is also analyzed on a real problem (Airfoil Shape Optimization). The impact of some ELMs parameters and an automatic model selection mechanism is investigated. The results obtained by ELMOEA/D are compared with those of two state-of-the-art surrogate approaches (MOEA/D-RBF and ParEGO) and a non-surrogate-based MOEA (MOEA/D). The ELMOEA/D variants are among the best results for most benchmark instances and for the real problem. HighlightsThe problem of solving expensive MOPs with small evaluation budgets is treated.The paper investigates the use of ELM as a surrogate model in MOEAs.A procedure is used to automatically select ELM parameters.The scalability of surrogates concerning the total of decision variables and number of objectives is evaluated.Experimental results show the good performance of ELMOEA/D in different problems.

Pedestrians, in videos taken from fixed cameras, tend to appear and disappear at precise locations such as doors, gateways or edges of the scene: we refer to locations where pedestrians appear as sources (potential origins) and the locations where they disappear as sinks (potential destinations). The detection of these points and the characterization of the flows connecting them represent a typical preliminary step in most pedestrian studies and it can be supported by computer vision approaches. In this paper we propose an algorithm in which a scene is overlaid by a grid of particles initializing a dynamical system defined by optical flow, a high level global motion information. Time integration of the dynamical system produces short particle trajectories (tracklets), representing dense but short motion patterns in segments of the scene; tracklets are then extended into longer tracks that are grouped using an unsupervised clustering algorithm, where the similarity is measured by the Longest Common Subsequence. The analysis of these clusters supports the identification of sources and sinks related to a single video segment. Local segment information is finally combined to achieve a global set of traces identifying sources and sinks, and characterizing the flow of pedestrians connecting them. The paper presents the defined technique and it discusses its application in a real-world scenario.

Facial expression-based emotion recognition has attracted lots of attention. Higher accurate performance could be expected with help of the other cues, e.g. physiological signals, for some specific blue such as 'Poker Face' and the lack of facial expression. In this paper, physiological signals are utilized for inferring user's emotion. However, the results of physiological-based emotion recognition are still inaccurate in the user-independent scenario since most existing methods ignore the difference in individual response pattern. To this end, we propose a Group-Based IRS (Individual Response Specificity) model to improve performance of physiological-based emotion recognition by taking user's IRS into account. The main contributions of this paper are two-fold: (1) an affective physiological database is collected to analyze human's emotional response pattern. The physiological signals are recorded from 30 subjects in four induced emotions (neutral, sadness, fear and pleasure). Three-channel bio-sensors are used to measure users electrocardiogram (ECG), galvanic skin response (GSR) and photo plethysmography (PPG). (2) In the experiment, the Group-based IRS model is proposed for emotion recognition in user-independent scenario, the effectiveness of which has been validated on our database. The results show that the Group-based IRS model can achieve higher recognition precision than the general model. HighlightsWe collect affective physiological dataset under four induced emotions.The theory of IRS is adopted to build a recognition model.We analyze and discuss the robustness and applicability of physiological signals.The Group-Based IRS model is utilized to improve performance of emotion recognition.