Skip to main content
main-content

Über dieses Buch

This book – in conjunction with the volumes LNCS 8588 and LNBI 8590 – constitutes the refereed proceedings of the 10th International Conference on Intelligent Computing, ICIC 2014, held in Taiyuan, China, in August 2014. The 85 papers of this volume were carefully reviewed and selected from numerous submissions. The papers are organized in topical sections such as soft computing; artificial bee colony algorithms; unsupervised learning; kernel methods and supporting vector machines; machine learning; fuzzy theory and algorithms; image processing; intelligent computing in computer vision; intelligent computing in communication networks; intelligent image/document retrievals; intelligent data analysis and prediction; intelligent agent and Web applications; intelligent fault diagnosis; knowledge representation/reasoning; knowledge discovery and data mining; natural language processing and computational linguistics; next gen sequencing and metagenomics; intelligent computing in scheduling and engineering optimization; advanced modeling, control and optimization techniques for complex engineering systems; complex networks and their applications; time series forecasting and analysis using artificial neural networks; computer human interaction using multiple visual cues and intelligent computing; biometric system and security for intelligent computing.

Inhaltsverzeichnis

Frontmatter

Soft Computing

Affective Tutoring System for Android Mobiles

Detecting and responding to affective states may be more influential than intelligence for tutoring success. This paper presents a software system that recognizes emotions of users using Android Cell Phones. The system software consists of a feature extractor, a neural network, and an intelligent tutoring system. The tutoring system, the neural network, and the emotion recognizer were implemented for running in Android devices. We also incorporate a novel fuzzy system, which is part of the intelligent tutoring system that takes actions depending of pedagogical and emotional states. The recognition rate of the emotion classifier was 96 %.

Ramón Zatarain-Cabada, M. L. Barrón-Estrada, José Luis Olivares Camacho, Carlos A. Reyes-García

Solving Three-Objective Flow Shop Problem with Fast Hypervolume-Based Local Search Algorithm

In this paper, we present a fast hypervolume-based multi-objective local search algorithm, where the fitness assignment is realized by the approximating computation of hypervolume contribution. In the algorithm, we define an approximate hypervolume contribution indicator as the selection mechanism and apply this indicator to an iterated local search. We carry out a range of experiments on three-objective flow shop problem. Experimental results indicate that our algorithm is highly effective in comparison with the algorithms based on the binary indicators and the exact hypervolume contribution indicator.

Rong-Qiang Zeng, Ming-Sheng Shang

Artificial Bee Colony Algorithms

A Learning Automata-Based Singular Value Decomposition and Its Application in Recommendation System

Recommendation system nowadays plays an important role in e-commerce, by helping consumers to find their preference from tens of thousands of goods and at the same time bringing large profits to e-commerce companies. Till now many different recommendation algorithm have been proposed and achieved good effect. In the context Netflix Prize in 2006, Simon Funk proposed a matrix factorization-based recommendation algorithm named Funk-SVD, which caused a widespread concern about the use of SVD model in recommend algorithm. Traditional SVD-based recommendation algorithm employs gradient descent algorithm as its optimization strategy. In this paper, we proposed a CALA-based algorithm to perform Funk-SVD, taking into consideration that CALA, as a kind of reinforcement learning model, has a superior performance on continues parameter optimization, especially in a unknown environment. As far as we known, the whole concept of CALA-based SVD is novel and unreported in the literature. To analyze the new algorithm, we tested it on the data set of film rating and achieved an average RMSE of 0.85, which is comparable with the former algorithm.

Yuchun Jing, Wen Jiang, Guiyang Su, Zhisheng Zhou, Yifan Wang

A Novel 2-Stage Combining Classifier Model with Stacking and Genetic Algorithm Based Feature Selection

This paper introduces a novel 2-stage classification system with stacking and genetic algorithm (GA) based feature selection. Specifically, Level1 data is first generated by stacking on the original data (called Level0 data) with base classifiers. Level1data is then classified by a second classifier (denoted by C) with feature selection using GA. The advantage of applying GA on Level1 data is that it has lower dimension and is more uniformity than Level0 data. We conduct experiments on both 18 UCI data files and CLEF2009 medical image database to demonstrate superior performance of our model in comparison with several popular combining algorithms.

Tien Thanh Nguyen, Alan Wee-Chung Liew, Xuan Cuong Pham, Mai Phuong Nguyen

Improved Bayesian Network Structure Learning with Node Ordering via K2 Algorithm

The precise construction of Bayesian network classifier from database is an NP-hard problem and still one of the most exciting challenges. K2 algorithm can reduce search space effectively, improve learning efficiency, but it requires the initial node ordering as input, which is very limited by the absence of the priori information. On the other hand, search process of K2 algorithm uses a greedy search strategy and solutions are easy to fall into local optimization. In this paper, we present an improved Bayesian network structure learning with node ordering via K2 algorithm. This algorithm generates an effective node ordering as input based on conditional mutual information. The K2 algorithm is also improved combining with Simulated Annealing algorithm in order to avoid falling into the local optimization. Experimental results over two benchmark networks Asia and Alarm show that this new improved algorithm has higher classification accuracy and better degree of data matching.

Zhongqiang Wei, Hongzhe Xu, Wen Li, Xiaolin Gui, Xiaozhou Wu

Combining Multi Classifiers Based on a Genetic Algorithm – A Gaussian Mixture Model Framework

Combining outputs from different classifiers to achieve high accuracy in classification task is one of the most active research areas in ensemble method. Although many state-of-art approaches have been introduced, no one method performs the best on all data sources. With the aim of introducing an effective classification model, we propose a Gaussian Mixture Model (GMM) based method that combines outputs of base classifiers (called meta-data or Level1 data) resulted from Stacking Algorithm. We further apply Genetic Algorithm (GA) to that data as a feature selection strategy to explore an optimal subset of Level1 data in which our GMM-based approach can achieve high accuracy. Experiments on 21 UCI Machine Learning Repository data files and CLEF2009 medical image database demonstrate the advantage of our framework compared with other well-known combining algorithms such as Decision Template, Multiple Response Linear Regression (MLR), SCANN and fixed combining rules as well as GMM-based approaches on original data.

Tien Thanh Nguyen, Alan Wee-Chung Liew, Minh Toan Tran, Mai Phuong Nguyen

Unsupervised Learning

A Precise Hard-Cut EM Algorithm for Mixtures of Gaussian Processes

The mixture of Gaussian processes (MGP) is a powerful framework for machine learning. However, its parameter learning or estimation is still a very challenging problem. In this paper, a precise hard-cut EM algorithm is proposed for learning the parameters of the MGP without any approximation in the derivation. It is demonstrated by the experimental results that our proposed hard-cut EM algorithm for MGP is feasible and even outperforms two available hard-cut EM algorithms.

Ziyi Chen, Jinwen Ma, Yatong Zhou

Comparison of EM-Based Algorithms and Image Segmentation Evaluation

Expectation-Maximization (EM) algorithm is used in statistics for finding maximum likelihood estimates of parameters in probabilistic models, where the model depends on unobserved latent variables. The idea behind the EM algorithm is intuitive and natural, which makes it applicable to a variety of problems. However, the EM algorithm does not guarantee convergence to the global maximum when there are multiple local maxima. In this paper, a random swap EM (RSEM) algorithm is introduced and compared to other variants of the EM algorithms. The variants are then applied to color image segmentation. In addition, a cluster validity criterion is proposed for evaluating the segmentation results from the EM variants. The purpose of this paper is to compare the characteristics of the variants with split and merge strategies and stochastic ways and their performance in color image segmentation. The experimental results indicate that the introduced RSEM performs better with simpler implementation than the other variants.

Mei Niu, Qinpei Zhao, Hongyu Li

Artificial Curiosity Driven Robots with Spatiotemporal Regularity Discovery Ability

Autonomous reinforcement learning (RL) robots usually need to learn from raw, high dimensional data generated by visual sensors and often corrupted by noise. These sorts of tasks are quite challenging and cannot be addressed without an efficient mechanism to encode and simplify the raw data. A recent study proposed an artificial curios robot (ACR) for this problem. However, this model is incapable of handling non-Markovian tasks and discovering spatiotemporal patterns in its milieu. This paper presents a method to solve this problem by extending ACR. A straightforward, but not efficient, solution is to keep recoding of previous observations which makes the algorithm intractable. We, instead, construct a perceptual context in a compact way. Using different environments, we show that the proposed algorithm can discover the regularity in its environment without any prior information on the task.

Davood Kalhor, Chu Kiong Loo

Kernel Methods and Supporting Vector Machines

The Equivalence Relationship between Kernel Functions Based on SVM and Four-Layer Functional Networks

This paper based on the concept of function interpolation, a functional network interpolation mechanism was analyzed, the equivalent between functional network and kernel functions based SVM, and the equivalent relationship between functional networks with SVM is demonstrated. This result provides us a very useful guideline when we perform theoretical research and applications on design SVM, functional network systems.

Yongquan Zhou, Qifang Luo, Mingzhi Ma, Liangliang Li

Multiple Kernel Learning Based on Cooperative Clustering

In recent years, kernel methods with single kernel had been challenged by the big data because of its heterogeneousness. In order to exploit the advantages of kernel methods, multiple kernel learning was proposed several years ago. However, the time and space complexity of multiple kernel learning increases greatly due to great amount computation of multiple kernels. So far, the research on improving training efficiency for multiple kernel learning mostly focuses on reducing the complexity of solving the objective function, other than reducing the training set. In this paper, the time performance of multiple kernel learning is improved through shrinking the training sets by introducing cooperative clustering, which is a novel method based on k-means clustering. Applying cooperative clustering to multiple kernel learning problems is proposed to reduce the number of support vectors, and then reduce the time complexity of multiple kernel learning algorithms. Experimental results show that the new method improves the efficiency of multiple kernel learning greatly with a slight impact on classification accuracy.

Haiyang Du, Chuanhuan Yin, Shaomin Mu

Machine Learning

Automatic Non-negative Matrix Factorization Clustering with Competitive Sparseness Constraints

Determination of the appropriate number of clusters is a big challenge for the bi-clustering method of the non-negative matrix factorization (NMF). The conventional determination method may be to test a number of candidates and select the optimal one with the best clustering performance. However, such strategy of repetition test is obviously time-consuming. In this paper, we propose a novel efficient algorithm called the automatic NMF clustering method with competitive sparseness constraints (autoNMF) which can perform the reasonable clustering without pre-assigning the exact number of clusters. It is demonstrated by the experiments that the autoNMF has been significantly improved on both clustering performance and computational efficiency.

Chenglin Liu, Jinwen Ma

Fusing Decision Trees Based on Genetic Programming for Classification of Microarray Datasets

In this paper, a genetic programming(GP) based new ensemble system is proposed, named as GPES. Decision tree is used as base classifier, and fused by GP with three voting methods: min, max and average. In this way, each individual of GP acts as an ensemble system. When the evolution process of GP ends, the final ensemble committee is selected from the last generation by a forward search algorithm. GPES is evaluated on microarray datasets, and results show that this ensemble system is competitive compared with other ensemble systems.

KunHong Liu, MuChenxuan Tong, ShuTong Xie, ZhiHao Zeng

A Study of Data Classification and Selection Techniques for Medical Decision Support Systems

Artificial Intelligence techniques have been increasingly used in medical decision support systems to aid physicians in their diagnosis procedures; making decisions more accurate and effective, minimizing medical errors, improving patient safety and reducing costs. Our research study indicates that it is difficult to compare different artificial intelligence techniques which are utilised to solve various medical decision-making problems using different data models. This makes it difficult to find out the most useful artificial intelligence technique among them. This paper proposes a classification approach that would facilitate the selection of an appropriate artificial intelligence technique to solve a particular medical decision making problem. This classification is based on observations of previous research studies.

Ahmed J. Aljaaf, Dhiya Al-Jumeily, Abir J. Hussain, David Lamb, Mohammed Al-Jumaily, Khaled Abdel-Aziz

A Reduction SVM Classification Algorithm Based on Adaptive AP Clustering Granulation

The classification speed of SVM is inversely proportional to the number of Support Vectors (SVs). Therefore, the less SVs means the more sparseness and the higher classification speed. In order to reduce the number of SVs but without losing of generalization performance, a new algorithm called Classification Algorithm of Support Vector Machine based on Adaptive Affinity Propagation clustering Granulation (CSVM-AAPG) is proposed, which employs Affinity Propagation (AP) clustering algorithm to cluster the original SVs and the cluster centers are used as the new SVs, then aiming to minimize the classification gap between SVM and CSVM-AAPG, a quadratic programming model is built for obtaining the optimal coefficients of the new SVs. Meanwhile, it is proven that when clustering the original SVs, the minimal upper bound of the error between the original decision function and the fast decision function can be achieved by AP. Finally, experiments show that compared with original SVs, the number of SVs decreases and the speed of classification increases using CSVM-AAPG, while the loss of accuracy is in the acceptable level.

Xiuxi Wei

Learning Automata Based Cooperative Student-Team in Tutorial-Like System

A novel learning automata (LA) based cooperative student-team in tutorial-like system is presented in this paper. The students in our system are modeled using LA. The new philosophy of a student is that he acquires knowledge not only from teacher, but also from team-workers. The self-examination indicator makes it possible for students to evaluate his learning outcomes. The below normal learner adopts the collective intelligence to improve himself. Experiments demonstrate the proposed method’s convergence speed is comparable to the student-classroom interaction method [9] and even better when the environment becomes harder. Compared to the previous interaction method and the single operated student, our accuracy is significantly improved.

Yifan Wang, Wen Jiang, Yinghua Ma, Hao Ge, Yuchun Jing

Clustering-Based Latent Variable Models for Monocular Non-rigid 3D Shape Recovery

The difficulty of monocular non-rigid 3D reconstruction using statistical learning approaches is to get a model that can represent as many deformations as possible. Given a known dataset to learn a model, existing latent variable models (LVMs) fail to focus on how to attain labeled samples. In this paper, we propose novel clustering-based LVMs in which we automatically select representative samples to be the labeled ones. To this end, G-means algorithm is adopted to cluster latent variables and obtain the labeled samples. These labeled samples are corresponding to the latent variables closest to clustering centers. We learn the Gaussian Process Latent Variable Model (GPLVM) and the Constrained Latent Variable Model (CLVM) into which we introduce clustering in the context of monocular non-rigid 3D reconstruction, and compare them to those without clustering. The experimental results show that our clustering-based LVMs could perform better.

Quan Wang, Fei Wang, Daming Li, Xuan Wang

Comments-Attached Chinese Microblog Sentiment Classification Based on Machine Learning Technology

Nowadays, with the rapid development of social networks, community-oriented Web sentiment analysis technology has gradually become a hot topic in the field of data mining. Being concise and flexible, Chinese microblog poses new challenges for sentiment analysis. This paper proposes an approach to classify Chinese microblog sentiments into positive and negative by the plain Naive Bayes (NB) and Support Vector Machine (SVM). Based on data preprocessing, sentiment lexicon construction, combining element of users’ comments, this research posit this Comments-attached Microblog Sentiment Classification, which is a novel method of attaching microblog users’ comments to the target microblog in order to improve the accuracy of sentiment classification. The experiment proves the vitality of this method and the advancement of the indecency from the way of language expressions.

Bo Yan, Bin Zhang, Hongyi Su, Hong Zheng

Evaluation of Resonance in Staff Selection through Multimedia Contents

In this paper we present the results of an experimental Italian research project finalized to support the classification process of the two behavioural status (resonance and dissonance) of a candidate applying for a job position. The proposed framework is based on an innovative system designed and implemented to extract and process the non-verbal expressions like facial, gestural and prosodic of the subject, acquired during the whole job interview session. In principle, we created our own database, containing multimedia data extracted, by different software modules, from video, audio and 3D sensor streams and then used SVM classifiers that perform in terms of accuracy 72%, 79% and 63% respectively for facial, vocal and gestural features. ANN classifiers have also been used, obtaining comparable results. Finally, we combined all the three domains and then reported the results of this last classification test proving that the experimental proposed work seems to perform in a very encouraging way.

Vitoantonio Bevilacqua, Angelo Antonio Salatino, Carlo Di Leo, Dario D’Ambruoso, Marco Suma, Donato Barone, Giacomo Tattoli, Domenico Campagna, Fabio Stroppa, Michele Pantaleo

Using Rough Set Theory and Decision Trees to Diagnose Enterprise Distress – Consideration of Corporate Governance Variables

This study discusses the key factors of financial distress warning models for companies using corporate governance variables and financial ratios as the research variables, sieving out influential variables based on the attribute simplification process of rough set theory (RST). Then, we construct some classification models for diagnosing enterprise distress based on RST, using a data mining technique of decision trees with the selected indicators and variables. The empirical results obtained from analysis of enterprise distress indicators, show that financial distress is not only affected by the traditional financial ratios, but also by corporate governance variables. In addition, enterprise distress diagnosis models constructed based on RST and decision trees can effectively diagnose firms in times of crisis. In particular, the RST models are more accurate. This study provides a reference for better understanding the symptoms that might lead to a company’s financial crisis in advance and thus provide a valuable reference for investment decision making by stakeholders.

Fu Hsiang Chen, Der-Jang Chi, Chun-Yi Kuo

Fuzzy Theory and Algorithms

Fuzzy Propositional Logic System and Its λ-Resolution

This paper researches resolution principle of the fuzzy propositional logic with contradictory negation, opposite negation and medium negation (FL

com

). In this paper, concepts of

λ

-satisfiable and

λ

-unsatisfiable are proposed under an infinite-valued semantic interpretation of FL

com

. The

λ

-resolution method of FL

com

is introduced. The

λ

-resolution deduction in FL

com

is defined and

λ

-resolution principle of FL

com

is discussed. Moreover, completeness theorem of the resolution method is proved.

Jiexin Zhao, Zhenghua Pan

Multi-level Linguistic Fuzzy Decision Network Hierarchical Structure Model for MCDM

Linguistic Fuzzy Decision network (LFDN) method is an extension of Fuzzy Decision Map (FDM) for solving Multi-Criteria Decision Making problems (MCDM) in fuzzy environment having dependence and feedback among criteria. On the other hand, LFDN can’t handle the complex decision making problem, particularly with the multi-level hierarchical structure model that consists of objectives, criteria, sub-criteria, etc. down to the bottom level (alternatives). The main objective of this paper is to develop the LFDN structure to be able to select a decision for multi-level structure problems. So the multi-level structure of LFDN is the general form of LFDN. Therefore, it can use for ranking alternatives and selecting the best one when the decision maker has multiple criteria. A case study was carried out to demonstrate the proposed model.

Basem Mohamed Elomda, Hesham Ahmed Hefny, Maryam Hazman, Hesham Ahmed Hassan

An Edge Sensing Fuzzy Local Information C-Means Clustering Algorithm for Image Segmentation

In this paper, we present a variation of fuzzy local information c-means (FLICM) algorithm for image segmentation by introducing a novel tradeoff factor and an effective kernel metric. The proposed tradeoff factor utilizes both local spatial and gray level information in a new way, and the Euclidean distance in FLICM algorithm is substituted by Gaussian Radial Basis function. By the novel factor and kernel metric, the new algorithm has edge identification ability and is insensitive to noise. Experiments result on both synthetic and real world images show that the proposed algorithm is effective and efficient, providing higher segmenting accuracy than other competitive algorithms.

Xinning Wang, Xiangbo Lin, Zhen Yuan

Application of Genetic Algorithm and Fuzzy Gantt Chart to Project Scheduling with Resource Constraints

Project scheduling with resource constraints is one of the most challenging optimization problems because of the complexity in estimating the resource requirement. This work aims at the application of fuzzy Gantt chart (FGC) and genetic algorithm (GA) to calculate optimal activity in project scheduling. Activity durations are considered adjustable for the optimal resource assignment under the constraints. GA determines not only the activity priority but also the activity duration within resource constraints. Numerical results of an example show that this application can effectively reduce the maximum resource input from 94 to 40 men with similar project makespan.

Yu-Chuan Liu, Hong-Mei Gao, Shih-Ming Yang, Chun-Yung Chuang

Dynamic Output Feedback Guaranteed Cost Control for T-S Fuzzy Systems with Uncertainties and Time Delays

This paper introduces a dynamic output feedback (DOF) guaranteed cost controller for systems with uncertainties and time delays, which are usually described by Takagi–Sugeno (T-S) fuzzy models. We model the nonlinear system directly by parallel distributed compensation (PDC) rules. The uncertainties and time delays caused by modeling errors and parameter sensitivity of devices are considered. Since states are not always available, we choose dynamic output feedback and design guaranteed cost controller based on linear matrix inequality (LMI). The upper bound of performances and calculation method for controller parameters are given. A numerical simulation is provided to demonstrate the availability and effectiveness of our method.

Guangfu Ma, Yanchao Sun, Jingjing Ma, Chuanjiang Li, Shuo Sun

Image Processing

Shape and Color Based Segmentation Using Level Set Framework

We propose a level set based variational approach that incorporates shape and color prior into Local Chan-Vese model for segmentation problem. Object detection and segmentation can be facilitated by the availability of a reference object. In our model, besides the level set function for segmentation, we introduce another labelling level set function to indicate the regions on which the prior shape and color should be compared. The active contour is able to find boundaries that are similar in shape and color to the prior, even when the entire boundary is not visible in the image. The experimental results demonstrate that the proposed model can efficiently segment the objects.

Xiang Gao, Ji-Xiang Du, Jing Wang, Chuan-Min Zhai

An Integrated NRSFM Approach for Image Sequences with Small Size

In this paper, a sub-sequence based integrated algorithm is proposed to deal with the non-rigid structure from motion (NRSFM) with small sequence size. In the proposed method, multiple sub-sequences are first extracted from the original sequence. Then, the extracted sub-sequences are used as the inputs of a recently reported NRSFM algorithm with rotation invariant kernel (RIK-NRSFM). Finally, the 3D coordinates estimated by the RIK-NRSFM algorithm are integrated to obtain the final estimation results. Experimental results on two widely used image sequences demonstrate the effectiveness and feasibility of the proposed algorithm.

Ya-Ping Wang, Zhan-Li Sun, Li Shang

Image Super-Resolution Reconstruction Based on Two-Stage Dictionary Learning

The general image super-resolution reconstruction (SRR) methods based on sparse representation utilizes the one-stage high and low resolution dictionary pairs to reconstruct a high resolution image, and this method can not restore much image detail information. To solve this detect, two-stage high and low resolution dictionaries are explored here. The goal of exploiting the two-stage dictionaries is to reconstruct the difference image between the original high resolution image and the reconstructed image obtained by using the one-stage dictionaries. In learning two-stage dictionaries, the difference image is used as the high resolution (HR) image, and the first-order and second-order gradient feature images of the one-stage reconstructed images are used as the low resolution (LR) images. Then, the two-stage dictionaries are learned by K- singular value decomposition (K-SVD) method. In test, an artificial and a real LR image are used, and simulation results show that, compared with other learning-based methods, our method proposed has remarkable improvement in PSNR and visual effect.

Li Shang, Zhan-li Sun

Intelligent Computing in Computer Vision

Position Accuracy Improvement of Robots having Closed-Chain Mechanisms

Improving the position accuracy of the closed chain robots needs a global robot kinematic model. Their closed loop constraints should be included in the global model via the specific passive joints of the robot. This paper proposes a simple but effective method to derive a mathematical formula for identification of all parameters of the robots having multiple closed chain mechanisms. An experimental calibration is carried out on a Hyundai HP160 robot to validate the correctness and effectiveness of the proposed method.

Hoai-Nhan Nguyen, Jian Zhou, Hee-Jun Kang, Tien-Dung Le

GPU-Based Real-Time Range Image Segmentation

In this paper proposed a GPU-based parallel processing method for real-time image segmentation with neural oscillator network. Range image segmentation methods can be divided into two categories: edge-based and region-based. Edge-base method is sensitive to noise and region-based method is hard to extracting the boundary detail between the object. However, by using LEGION (Locally Excitatory Globally Inhibitory oscillator networks) to do range image segmentation can overcome above disadvantages. The reason why LEGION is suitable for parallel processing that each oscillator calculate with its 8-neiborhood oscillators in real time when we process image segmentation by LEGION. Thus, using GPU-based parallel processing with LEGION can improve the speed to realize real-time image segmentation.

Xinhua Jin, Dong Joong Kang, Mun-Ho Jeong

Recognition of Human Action and Identification Based on SIFT and Watermark

This paper presents a fast and simple method for action recognition and identity at the same time. A watermark embedding as a 2-D wavelet in the training data at the first step to identify the identity of who makes the action. The proposed technique relies on detecting interest points using SIFT (scale invariant feature transform) from each frame of the video for action recognition. More specifically, we propose an action representation based on computing a rich set of descriptors from 2D-SIFT key points. Since most previous approaches to human action recognition typically focus on action classification or localization, these approaches usually ignore the information about human identity. A compact yet discriminative semantics visual vocabulary was built by a K-means for high-level representation. Finally a multi class linear Support Vector Machine (SVM) is utilized for classification. Our algorithm can not only categorize human actions contained in the video, but also verify the person who performs the action. We test our algorithm on three datasets: the KTH human motion dataset, Weizmann and our action dataset. Our results reflect the promise of our approach.

Khawlah Hussein Ali, Tianjiang Wang

Augmented Reality Surveillance System for Road Traffic Monitoring

This paper introduces the augmented reality surveillance system which evaluates the density of the traffic on roads and displays information in an easy to understand form over the video stream and a map. A mutual dependence between the real world, global coordinates and the position of the pixel in the image is explained. The way to find the real size of an object by knowing its dimension in the image is introduced. An operator can decide what points on the map it is required to survey, and the camera will know how to rotate to those points by mapping of global coordinates to pan and tilt angles. The density of the traffic is evaluated by processing video data and applying the knowledge about real width and length of cars.

Alexander Filonenko, Andrey Vavilin, Taeho Kim, Kang-Hyun Jo

Effective Palm Tracking with Integrated Tracker and Offline Detector

In this paper, we propose a vision based palm tracking method with three inherently connected components: i) an offline palm detector that locates all possible palm-like objects; ii) a SURF-based tracking module that identifies the tracked palm’s location using historical information; iii) an adaptive skin color model and a patch similarity calculation module. The outputs from the last component can effectively eliminate false detections and decide which palm is under tracking and also provide updated information to the first two modules. In summary, our work makes the following contributions: i) an effective offline palm detector; ii) a benchmark dataset for training and testing palm detectors; iii) an effective solution to tackling the challenges of palm tracking in adverse environments including occlusions, changing illumination and lack of context. Experiment results show that our method compare favorably with other popular tracking techniques such as Camshift and TLD in terms of precision and recall.

Zhibo Yang, Yanmin Zhu, Bo Yuan

Monocular 3D Shape Recovery of Inextensibility Deformable Surface by Using DE-Based Niching Algorithm with Partial Reinitialization

Template-based deformable surface shape recovery is a well-known challenging problem for its compatible local minima and high degree of freedom. The gradient-based optimization method often converges to the local minimum, the premature convergence also occurs even using the evolution strategies which are highly effective in locating a single global minimum. Meanwhile, exploration in a high dimensional space is often time exhausted. To avoid these difficulties, a two-step method was proposed. The projections of vertices of a mesh were estimated firstly. Then the 3D positions of the vertices were estimated via estimating the depth along the sightlines calculated according to the given projections. While the depth of vertices was estimated, the problem was regarded as a multimodal optimization. A DE-based niching algorithm was used to solve it, and the partial reinitialization was used to keep the diversity of the population. The effectiveness of our method was demonstrated on both synthetic data and real images.

Xuan Wang, Fei Wang, Lei Chen

Computer Vision Based Traffic Monitoring System for Multi-track Freeways

Nowadays, development is synonymous with construction of infrastructure. Such road infrastructure needs constant attention in terms of traffic monitoring as even a single disaster on a major artery will disrupt the way of life. Humans cannot be expected to monitor these massive infrastructures over 24/7 and computer vision is increasingly being used to develop automated strategies to notify the human observers of any impending slowdowns and traffic bottlenecks. However, due to extreme costs associated with the current state of the art computer vision based networked monitoring systems, innovative computer vision based systems can be developed which are standalone and efficient in analyzing the traffic flow and tracking vehicles for speed detection. In this article, a traffic monitoring system is suggested that counts vehicles and tracks their speeds in realtime for multi-track freeways in Australia. Proposed algorithm uses Gaussian mixture model for detection of foreground and is capable of tracking the vehicle trajectory and extracts the useful traffic information for vehicle counting. This stationary surveillance system uses a fixed position overhead camera to monitor traffic.

Zubair Iftikhar, Prashan Dissanayake, Peter Vial

Robust Pose Estimation Algorithm for Approximate Coplanar Targets

To uniquely determine the position and orientation of a calibrated camera from a single image, pose estimation algorithms have been developed. However, the presented algorithms usually encounter pose ambiguity problem when process approximate coplanar targets, which can be defined as that a majority of object points on these targets belongs to a plane while some others distract outside the plane. Based on a more comprehensive explanation for pose ambiguity from the influence of 3D object configuration, we propose a robust pose estimation algorithm. The approximate coplanar points are divided into coplanar points and non-coplanar points. When two candidate solutions are calculated by coplanar points, final pose is determined using non-coplanar points. Simulation results and experiments on real images prove the effectiveness of our proposed pose estimation algorithm.

Haiwei Yang, Fei Wang, Lei Chen, Yicong He, Yongjian He

A BP Neural Network Predictor Model for Stock Price

This paper presents a stock price forecast model using LM-BP algorithm, which has adopted three-layer back propagation neural networks. This model is more fast convergence rate and overcomes redundancy and noise of the samples. The proposed model is simulated in MATLAB platform and the experimental results of stocks price prediction show that the LM-BP model attains high accuracy of predicting the stock price in short-term. This paper provides also the comparison of actual value and forecast value, which proves that the model is effective and promising.

Li Huo, Bo Jiang, Tao Ning, Bo Yin

Intelligent Computing in Communication Networks

Plant Leaf Recognition Using Histograms of Oriented Gradients

Leaves from plants are proved to be a feasible source of information used to identify plant species [1]. In this paper, we present a method to recognize plant leaves employing Histograms of Oriented Gradients (HOG) as the feature descriptor. For better robustness to illumination, shadow, quality degradation, etc., five vital factors of original HOG algorithm are discussed to evaluate the respective effects in different configurations. Experimental results show that this method achieves excellent performance in recognition rate.

Qing Xia, Hao-Dong Zhu, Yong Gan, Li Shang

An Implementation of an Intelligent PCE-Agent-Based Multi-domain Optical Network Architecture

This paper presents a new optical network architecture based on the PCE (Path Computation Element) by adding a PCE-Agent to intelligently control PCEs in each layer of the conventional multi-domain optical network. The PCE-Agent uses a weighted sum policy to evaluate the performance of each domain managed by a PCE in each layer for the selection of the next hop domain when establishing a P2P (Point to Point) or P2MP (Point to Multiple Points) routing path. A PA-BRPC (PCE-Agent Backward Recursive PCE-based Computation Algorithm) routing algorithm is then proposed by extending the BRPC (Backward Recursive PCE-based Computation Algorithm) algorithm in the PCE-Agent-based multi-domain optical network architecture. Compared with the traditional BRPC algorithm, our PA-BRPC algorithm based on the improved PCE-Agent architecture obtains better performance in terms of both the utilization rate of resources and the load balance of the optical network.

Ying Xu, Tiantian Zhang, Renfa Li

A Heuristic Virtual Network Mapping Algorithm

Virtual Network Mapping Problem (VNMP) is one of the key problems in network virtualization, which is proved as a non-deterministic polynomial hard (NP-Hard) problem. Considering the relevance of nodes and links, a virtual network mapping algorithm based on biogeography optimization algorithm is proposed in this paper. In order to reduce the cost of the substrate network embedding, it is designed as a one-stage algorithm and solved with heuristic algorithm. The experimental results suggest that the proposed algorithm increases the average operating income of the virtual networks and reduces the cost of substrate network comparing with other two-stage mapping algorithms.

Xiao-guang Wang, Xiang-wei Zheng, Dian-jie Lu

Fault-Tolerant Control, Fault Diagnosis and Recovery in Runtime of Business Docking Service Composition Flow in the Cloud Environment

In the design and remote maintenance of district and city petition business docking platform, loose coupling environment of service composition flow, and autonomy of distributed resources can easily lead to failures of inter-process interaction, the platform running will gradually changes form trouble-free operation to operation with fault, so we design a docking platform with fault-isolation and fault-tolerance control based on services choreography and orchestration, and a remote monitoring system based on fault-diagnosis by fussy inference. More than two years’ operation and remote maintenance verify the effectiveness of the fault-tolerant control mode in the process of business docking services composition. Achieve the goal, that the improving of fault-recovery mechanism and the changing of business docking requirements won’t affect each other. Accordingly, it reduces the complexity of maintenancing services composition software, provides application value in business docking service mode for service computing and cloud computing, and SaaS-level monitoring and repair method.

Jianhua Han, Silu He, Hengxin Li, Jianping Huang

Intelligent Image/Document Retrievals

Ambiguous Proximity Distribution

Proximity Distribution Kernel is an effective method for bag-of-featues based image representation. In this paper, we investigate the soft assignment of visual words to image features for proximity distribution. Visual word contribution function is proposed to model ambiguous proximity distributions. Three ambiguous proximity distributions is developed by three ambiguous contribution functions. The experiments are conducted on both classification and retrieval of medical image data sets. The results show that the performance of the proposed methods, Proximity Distribution Kernel (PDK), is better or comparable to the state-of-the-art bag-of-features based image representation methods.

Quanquan Wang, Yongping Li

Indexing SURF Features by SVD Based Basis on GPU with Multi-Query Support

Space partitioning based indexing technique reduces search space and increases performance of the retrieval system. Geometric hashing is one such space partitioning technique which creates a global model descriptor. Aligning points prior to geometric hashing proves to be better for increasing the retrieval performance. This paper presents a newer technique of aligning points of Speeded Up Robust Features (SURF) using basis direction obtained through Singular Value Decomposition (SVD). Performance gets boosted by 10% on average in online searching as compared to classical approach. Further, the implicit parallelism that exist in proposed technique motivated us to use Graphics Processing Unit (GPU). The GPU based implementation of the proposed indexing technique achieved speed up in the range of 14.3x-82.28x for offline database indexing and 3.54x-4255.88x for online searching. Extension to the single query execution is provided by the newer multi-query approach. It proves to be better than the linear execution of multi-query. For GPU based multi-query implementation, speed up obtained is in the range of 3.74x-3097.55x for 1 to 10 queries to be executed simultaneously.

Vibha Patel, Bhavin Patel

Intelligent Data Analysis and Prediction

Dynamically Changing Service Level Agreements (SLAs) Management in Cloud Computing

Managing Service Level Agreements (SLAs) in cloud computing in a dynamic manner becomes a critical issue for cloud service providers. This is due to the emerging technology and the frequent and continuous change in cloud service requirements and techniques over the time. Current cloud SLAs management methods are rigorous in terms of SLA updation to incorporate new changes. Since updating any SLA, to meet any required change in cloud requirements and techniques, requires reformulation and remapping to all available SLAs (called public SLAs). This paper proposed a mechanism to dynamically manage cloud computing SLAs based on Real Options Analysis (ROA) concept. The proposed model maps the required changes to all public SLAs and sorts out the most related or suitable SLAs (solutions) based on options theory while recording the other solutions for any future change according to emerging circumstances. The technique incorporates any new change dynamically; by mapping it to a limited number of SLAs (recorded solutions) based on various options presented by ROA. The framework presented in this paper would provide a flexible solution in managing cloud SLAs in both cloud provider and the user’s perspectives.

Waleed Halboob, Haider Abbas, Kamel Haouam, Asif Yaseen

Foundry Material Design with Artificial Intelligence

There are two main development trends in the modeling techniques for the manufacturing procedure of foundry material. One is based on numerical simulation, the other is based on artificial intelligence. The numerical simulation method depend on the strict mathematics model and scientific mechanism, therefore, it is difficult to predict the final structure and properties by computers. The artificial intelligence method can learn from the empirical data, summarize regularity, automatically build models, and predict the future as human brain does. Focusing on the complexity of foundry material design and combining advanced research in the field of artificial intelligence, this paper describes the application and development orientation about artificial intelligence technology in foundry material design, expounds the features of various technologies. In particular, the applications of artificial intelligence in some way of foundry material properties predicting are summarized.

Jingjing Zhao, Xingtong Liu, Afeng Yang, Chun Du

Intelligent Agent and Web Applications

An Intelligent Agent Simulation Model to Evaluate Herd Behavior and Sales Effort in a Duopoly Market

This paper studies the impact of ‘herd behavior’ consumers in the market on demand of the products. In the duopoly market, two suppliers locate at two end of the line separately, providing an undifferentiated product. The potential consumers are dispersed along the line randomly. Beside those ‘rational’ consumers who have willing to buy from one of the two suppliers according to transportation cost from their positions to the two suppliers, other ‘herd behavior’ consumers may ignore the transportation cost, and change their mind to follow the most popular choice. To represent herd behavior of heterogeneous consumers, multi-agent based modeling and simulation is introduced to model and testify the operations of the market. Our numerical results show that ‘herd behavior’ consumers have a great influence on demand of two suppliers. Moreover, demand can be obviously increased once the first coming consumers are influenced by the supplier through sales efforts.

Feng Li, Ying Wei

A “Content-Behavior” Learner Model for Adaptive Learning System

The learner model in adaptive learning system plays an important role. The study aims to explore the main characteristics and components during students’ engaging in online learning system, proposing a “content-behavior” learner model based on combining students’ learning behavior with the grasp degree towards concepts. Especially, the learner model should be open to students in order to make them look into their learning process. And further, they can review peer’s learning portfolio and then reflect their learning. The result of this study should provide some suggestions towards the research in the fields of adaptive eLearning in online learning environment.

Qingchun Hu, Yong Huang, Yi Li

Aggregate MAC Based Authentication for Secure Data Aggregation in Wireless Sensor Networks

Wireless sensor networks perform in-network processing to reduce the energy consumption caused by redundant communication. At the same time, its hostile deployment and unreliable communication raise the security concerns. Thus, there is a need to blend security and data aggregation together to provide secure data aggregation. Secure data aggregation becomes challenging if end-to-end privacy is desired. Privacy homomorphism is used to achieve both en route aggregation and end-to-end privacy of sensor readings. However, privacy homomorphism is inherently malleable. Using privacy homomorphism, one can modify the ciphertext without decrypting it. Thus, it becomes extremely crucial to ensure authentication along with privacy. Symmetric key based Message Authentication Code (MAC) is an efficient solution to provide authentication. In this paper, we use Aggregate Message Authentication Codes (AMAC) to reduce the transmission cost incurred by MAC. However, conflicting requirements of AMAC and data aggregation make its usage limited for certain scenarios. In this paper, we present a cluster based scenario where we can apply AMAC to reduce the number of bits transmitted for authentication.

Keyur Parmar, Devesh C. Jinwala

Intelligent Fault Diagnosis

Active Learning Methods for Classification of Hyperspectral Remote Sensing Image

Active learning(AL) is an effective method in definition of samples, especially when labeled sample number is small. In this paper, we propose two active learning algorithms, which are Random Sampling (RS) and Margin Sampling(MS) algorithms, the two techniques achieve semiautomatic definition of training samples in remote sensing image classification, starting with a small and representative data set, then according to query criterion, the experts select informative samples to add training data set, the model builds the optimal set of samples which minimizes the classification error. Compared with traditional sample selection methods, the results denote the effectiveness of the proposed AL methods.

Sheng Ding, Bo Li, Xiaowei Fu

Time to Fault Minimization for Induction Motors Using Wavelet Transform

Time to Fault (TTF) is an important parameter that measures how long it takes that a fault detection algorithm successfully recognizes defect in the motor. If TTF is too long, severe damages can happen to the motor. In this paper, authors try to minimize TTF using Discrete Wavelet Transform (DWT); in other words, the output signals derived from the motor due to an existing fault are analyzed and decomposed in frequency-domain. It will be proved that even though there are

n

levels for decomposing the signal with

2

n

data samples, but after a specific level, the fault characteristics will disappear. This happens because of sporadic form of the signal. Thus, we can finish the analysis in a lower level where all characteristics for fault can be seen. This reduces TTF and consequently possible damages considerably.

Amirhossein Ghods, Hong-Hee Lee

Knowledge Representation/Reasoning

Legal Reasoning Engine for Civil Court Procedure

In this paper, we propose a new inference engine for legal reasoning in civil court (LR engine). Three main modules of the engine consist of the Prove algorithm, the Defense algorithm and the Estoppel algorithm. These modules work in switching style in order to drive the legal reasoning according to the Evidence law and the doctrine of Estoppel. In this study, Thai Civil and Commercial Code Law are applied as a legal knowledge base. Several legal cases are examined to validate LR engine and analyze its performance. The experimental result shows that LR engine can automatically handle the case containing estoppel and can reduce the steps of proof. Moreover, it provides reasonable legal explanation compared to the traditional algorithm.

Tanapon Tantisripreecha, Ken Satoh, Nuanwan Soonthornphisaj

A Multimodal Fingers Classification for General Interactive Surfaces

In this paper a multimodal fingers classification to detect touch points over general interactive surfaces is presented. Three different classifiers have been used: artificial neural networks, decision trees and rules learner. The data set has been created extracting statistical parameters from finger ROIs on about 40000 video samples. The accuracy obtained for the three classifiers on the test set is respectively 96,68%, 96,58% and 97,41%. The model classifiers generated work very well in real-time applications, so an innovative software called TouchPAD has been designed and implemented.

Vitoantonio Bevilacqua, Donato Barone, Marco Suma

A Signal Modulation Type Recognition Method Based on Kernel PCA and Random Forest in Cognitive Network

This paper develops a solution for the problem of the low accuracy on signal modulation type recognition of the weak primary users in low signal-to-noise ratio, a novel signal recognition method based on dimensionality reduction and random forest (RF) is proposed. Firstly, the kernel principal component analysis (KPCA) is applied to extract the most discriminate feature vector. Secondly, the detecting signal is classified by the trained random forest. Performance evaluation is conducted through simulations, and the results reveal the benefits of adopting the proposed algorithm comparing with support vector machine (SVM) and PCA-SVM algorithms.

Xin Wang, Zhijun Gao, Yanhui Fang, Shuai Yuan, Haoxuan Zhao, Wei Gong, Minghao Qiu, Qiang Liu

Knowledge Discovery and Data Mining

Privacy-Preserving Data Mining Algorithm Based on Modified Particle Swarm Optimization

The privacy preserving data mining is a research hotspot. Most of the privacy preserving algorithms are focused on the centralized database. The algorithms on the distributed database are very vulnerable to collusion attack. The Privacy-Preserving data mining algorithm based on particle swarm optimization is proposed in this paper. The algorithm is based on centralized database, and it can be used on the distributed database. The algorithm is divided into two steps in the distributed database. In the first step, the modified particle swarm optimization algorithm is used to get the local Bayesian network structure. The purpose of the second step is getting the global Bayesian network structure by using local ones. In order to protect the data privacy, the secure sum is used in the algorithm. The algorithm is proved to be convergent on theory. Some experiments have been done on the algorithm, and the results prove that the algorithm is feasible.

Lei Yang, Jue Wu, Lingxi Peng, Feng Liu

Topic Extraction Based on Knowledge Cluster in the Field of Micro-blog

Views in the field of Micro-blog can always express what people concerns. Topic extraction in this domain is necessary to help focus on events that are popular and important. At present LDA, a probabilistic topic model based on word co-occurrence is widely used. However, when it comes to deal with microbloggings that are all short texts with a lot of noise, the result can be not satisfactory. In this paper, we propose a new index called M-membership that is proved to be more suitable to indicate the importance of each term in this scenario, and present a topic extraction method based on Fuzzy C-Means algorithm, then use fuzzy set to represent clusters and topics, which can lead to a more reasonable and thematic result. Comparative experiments with K-Means and LDA illustrate our ability to cluster and extract topics in the field of Micro-blog.

Ming Li, Chunhong Zhang, Li Sun, Xianlei Shao

Multi-strategy Based Sina Microblog Data Acquisition for Opinion Mining

As an important media for social interactions and information dissemination through the internet, Sina microblog contains emotional state and important opinion of participants. Dealing with microblog data belongs to big data areas, the premise of which is to obtain a large amount of microblog data for further analysis and data mining. For commercial interests as well as security considerations, the access to the data is becoming increasingly difficult and the API Sina microblog officially provided doesn’t support large amount of data mining. In this paper, we try to design a platform that is mainly based on the access mechanism of multi-strategy and existing resources to collect data stably from Sina microblog. The results demonstrate that a combination of API and web crawler allows efficient data mining. In such way, sentiment analysis and opinion mining are performed on the data obtained by the multi-strategy method, which proved that the proposed solutions will be allowed to build straightforward application of hot words searching, opinion mining and sentiment analysis.

Xiao Sun, Jia-qi Ye, Fu-ji Ren

Mining Longest Frequent Patterns with Low-Support Items

Finding the longest sequential pattern is a basic task in data mining. Current algorithm often depends on the cascading support counting, including the famous apriori algorithm, FP-growth, and some likewise derived algorithms. One thing should be pointed out that in these sorts of algorithms the items with very high support may lead to a poor time performance and very huge useless search space, especially when the items in fact are not the member of the result longest pattern. We reconsider the role of connection between the items and carefully analysis the hidden chains connecting items. Thus a new method of mining the longest pattern in transaction database is proposed in this paper. Our algorithm can have better performance when overcoming the bad side-effect of big-support items, especially in case of the items being not members of the result longest pattern.

Qinhua Huang, Weimin Ouyang

Integrating Time Stamps into Discovering the Places of Interest

With the employment of GPS embedded device, large numbers of data has been collected from location aware applications. It is interesting and challenging to discover meaningful information behind the data. Since the GPS data contains the time information, we take use of the time stamps of the GPS data in this paper for better discovering the places of interest. The collection usually contains large amounts of trajectories, where not every point has information. Therefore, a time stamp clustering algorithm is firstly proposed to reduce the size of raw data and also extract the points with more information. Different clustering algorithms are then conducted on the pre-processed data for extracting the places of interest. Finally, we compare the clustering algorithms on the GPS data by several external validity indexes.

Jun Zhou, Qinpei Zhao, Hongyu Li

A Hybrid Solution of Mining Frequent Itemsets from Uncertain Database

With the emergence of new applications, the traditional way of mining frequent itemsets is not available in uncertain environment. In the past few years, researchers presented different solutions in extending conventional algorithms into uncertainty environment. In this paper, we review previous algorithms and proposed a hybrid solution to mine frequent itemsets from uncertain databases. The new scheme bases on traditional Eclat algorithm and mines frequent itemsets under the definition of frequent probability. Furthermore, the hybrid solution exerts fuzzy mining and precise mining adaptively according to the characters of the candidate databases, which addresses the problem of tradeoff between computation and accuracy. We tested our solution on a number of uncertain data sets, and compared it with the well known uncertain frequent itemsets mining algorithms. The experimental results show that our solution is efficient and accuracy.

Xiaomei Yu, Hong Wang, Xiangwei Zheng

DVT-PKM: An Improved GPU Based Parallel K-Means Algorithm

K-Means clustering algorithm is a typical partition-based clustering algorithm. Its two major disadvantages lie in the facts that the algorithm is sensitive to initial cluster centers and the outliers exert significant influence on the clustering results. In addition, K-Means algorithm traverses and computes all the data multiple times. Thus, the algorithm is not efficient when dealing with large data sets. In order to overcome the above limitations, this paper proposes to exclude the outliers using the minimum number of points in the d-dimensional hypersphere area. Then k cluster centers can be obtained by adjusting the threshold making use of density idea. Finally, K-Means algorithm will be integrated with Compute Unified Device Architecture (CUDA). The time efficiency is improved considerably through taking advantage of computing power of Graphic Processing Unit (GPU). We use the ratio of distance between classes to distance within classes and speedup as the evaluation criteria. The experiments indicate that the proposed algorithm significantly improves the stability and running efficiency of K-Means algorithm.

Bo Yan, Ye Zhang, Zijiang Yang, Hongyi Su, Hong Zheng

A Multi-Intelligent Agent for Knowledge Discovery in Database (MIAKDD): Cooperative Approach with Domain Expert for Rules Extraction

In last decade, autonomous intelligent agents or multi-intelligent agents and knowledge discovery in database are combined to produce a new research area in intelligent information technology. In this paper, we aim to produce a knowledge discovery approach to extract a set of rules from a dataset for automatic knowledge base construction using cooperative approach between a multi-intelligent agent system and a domain expert in a particular domain. The proposed system consists of several intelligent agents, each one has a specific task. The main task is assign to associative classification mining intelligent agent to deal with a database directly for rules extraction using Classification Based on Associations (CBA) rule generation and classification algorithm, and send them to a domain expert for a modification process. Then, the modified rules will be saved in a knowledge base which is used later by other systems (e.g. knowledge-based system). In other words, the aim of this work is to introduce a tool for extracting knowledge from database, more precisely this work has focused on produce the knowledge base automatically that used rules approach for knowledge representation. The MIAKDD is developed and implemented using visual Prolog programming language ver. 7.1 and the approach is tested for a UCI heart diseases dataset.

Mohammed Abbas Kadhim, M. Afshar Alam, Harleen Kaur

Natural Language Processing and Computational Linguistics

The Role of Pre-processing in Twitter Sentiment Analysis

Recently, increasing attention has been attracted to Social Networking Sentiment Analysis. Twitter as one of the most fashional social networking platforms has been researched as a hot topic in this domain. Normally, sentiment analysis is regarded as a classification problem. Training a classifier with tweets data, there is a large amount of noise due to tweets’ shortness, marks, irregular words etc. In this work we explore the impact pre-processing methods make on twitter sentiment classification. We evaluate the effects of URLs, negation, repeated letters, stemming and lemmatization. Experimental results on the Stanford Twitter Sentiment Dataset show that sentiment classification accuracy rises when URLs features reservation, negation transformation and repeated letters normalization are employed while descends when stemming and lemmatization are applied. Moreover, we get a better result by augmenting the original feature space with bigram and emotions features. Comprehensive application of these measures makes us achieve classification accuracy of 85.5%.

Yanwei Bao, Changqin Quan, Lijuan Wang, Fuji Ren

Construction of a Chinese Emotion Lexicon from Ren-CECps

This paper presents an automatic method to build a Chinese emotion lexicon based on the emotion corpus Ren-CECps. The method includes word extraction and emotion classification. Firstly, sentences are parsed to extract candidate emotional words. By making use of the words co-occurrence in the corpus, we get the similarity between words. And then Support Vector Machine (SVM) is adopted to classify the candidate emotional words. Experiment on the manual labeled words has shown that our classification method achieved high precision. Finally we apply our method on unlabeled corpus to get emotional words.

Lijuan Wang, Changqin Quan, Yanwei Bao, Fuji Ren

Word Frequency Statistics Model for Chinese Base Noun Phrase Identification

The Chinese base phrase identification plays an important role in the field of natural language processing. It needs to be improved in the recognition scope and methods currently. This paper presents a method based on word frequency statistics model for Chinese base noun phrase identification: Building the noun phrase dictionary by training corpus, calculating the co-occurrence frequency and threshold of the noun phrase, and constructing word table according to the different roles of the words in the noun phrase. Unknown word processing and rule templates are added. Improve the results with error correction processing at last. Experiments on the test corpus show that the average precision and average recall rate of the base noun phrases identification in different areas are 91.28% and 93.22%.

Lu Kong, Fuji Ren, Xiao Sun, Changqin Quan

Structure Constrained Discriminative Non-negative Matrix Factorization for Feature Extraction

In this paper, we propose a novel algorithm called Structure Constrained Discriminative Non-negative Matrix Factorization (SCDNMF) for feature extraction. In our proposed algorithm, a pixel dispersion penalty (PDP) constraint is employed to preserve spatial locality structured information of the basis obtained by NMF. At the same time, in order to improve the classification performance, intra-class graph and inter-class graph are also constructed to exploit discriminative information as well as geometric structure of the highdimensional data. Therefore, the low-dimensional features obtained by our algorithm are structured sparse and discriminative. Moreover, an iterative updating optimization scheme is developed to solve the objective function of the proposed SCDNMF. The proposed method is applied to the problem of image recognition using the well-known ORL, Yale and COIL20 databases. The experimental results demonstrate that the performance of our proposed SCDNMF outperforms the state-of-the-art methods.

Yan Jin, Lisi Wei, Yugen Yi, Jianzhong Wang

Next-Gen Sequencing and Metagenomics

Simulated Annealing Based Algorithm for Mutated Driver Pathways Detecting

With the development of Next-generation DNA sequencing technologies, one of the challenges is to distinguish functional mutations vital for cancer development, and filter out the unfunctional and random “passenger mutations.” In this study, we introduce a modified method to solve the so-called maximum weight submatrix problem which is based on two combinatorial properties, i.e., coverage and exclusivity. This problem can be used to find driver mutations. Particularly, we enhance an integrative model which combines mutation and expression data. We apply our method to simulated data, the experiment shows that our method is efficiency. Then we apply our proposed method onto real biological datasets, the results also show that it is applicable in real applications.

Chao Yan, Hai-Tao Li, Ai-Xin Guo, Wen Sha, Chun-Hou Zheng

Special Session on Intelligent Computing in Scheduling and Engineering Optimization

Grouped Fruit-Fly Optimization Algorithm for the No-Wait Lot Streaming Flow Shop Scheduling

Lot streaming involves splitting products into several sublots to enhance the processing flexibility. In this paper the no-wait lot streaming flow shop scheduling problem is discussed with the constraint of unintermingling consistent-size sublots. The number of sublots, sublots sizes and the sequence of products are optimized by a proposed grouped fruit fly optimization algorithm (GFOA). We design three basic neighborhood structures and a structure based on the no-wait property to generate new solutions. A speed-up evaluation method is presented to improve the efficiency. Numerical test results and comparisons are provided, which demonstrate the effectiveness of the proposed GFOA.

Peng Zhang, Ling Wang

An Enhanced Estimation of Distribution Algorithm for No-Wait Job Shop Scheduling Problem with Makespan Criterion

In this paper, an enhanced estimation of distribution algorithm (EEDA) is proposed for the no-wait job shop scheduling problem (NWJSSP) with the makespan criterion, which has been proved to be strongly NP-hard. The NWJSSP can be decomposed into the sequencing and the timetabling problems. The proposed EEDA and a shift timetabling method are used to address the sequencing problem and the timetabling problem, respectively. In EEDA, the EDA-based search is applied to guiding the search to some promising sequences or regions, and an Interchange-based local search is presented to perform the search from these promising regions. Moreover, each individual or sequence of EEDA is decoded by applying a shift timetabling method to solving the corresponding timetabling problem. The experimental results show that the combination of the EEDA and the shift timetabling method can accelerate the convergence speed and is helpful in achieving more competitive results.

Shao-Feng Chen, Bin Qian, Rong Hu, Zuo-Cheng Li

Bayesian Statistical Inference-Based Estimation of Distribution Algorithm for the Re-entrant Job-Shop Scheduling Problem with Sequence-Dependent Setup Times

In this paper, a bayesian statistical inference-based estimation of distribution algorithm (BEDA) is proposed for the re-entrant job-shop scheduling problem with sequence-dependent setup times (RJSSPST) to minimize the maximum completion time (i.e., makespan), which is a typical NP hard combinatorial problem with strong engineering background. Bayesian statistical inference (BSI) is utilized to extract sub-sequence information from high quality individuals of the current population and determine the parameters of BEDA’s probabilistic model (BEDA_PM). In the proposed BEDA, BEDA_PM is used to generate new population and guide the search to find promising sequences or regions in the solution space. Simulation experiments and comparisons demonstrate the effectiveness of the proposed BEDA.

Shao-Feng Chen, Bin Qian, Bo Liu, Rong Hu, Chang-Sheng Zhang

An Effective Estimation of Distribution Algorithm for Multi-track Train Scheduling Problem

In this paper, an effective estimation of distribution algorithm (EDA) is presented for solving the multi-track train scheduling problem (MTTSP). The individual of the EDA is represented as the permutation of train priority. With a proper track assignment rule, the individual is decoded into feasible schedule. In addition, the EDA builds a probability model for describing the distribution of the solution space. In every generation, it samples the promising region for generating new individuals and updates the probability model with the superior population. Moreover, the influence of parameter setting is investigated based on design-of-experiment method and a set of suitable parameter values is suggested. Simulation results based on some instances and comparisons with the existing algorithm demonstrate the effectiveness and efficiency of the EDA.

Shengyao Wang, Ling Wang

Special Session on Advanced Modeling, Control and Optimization Techniques for Complex Engineering Systems

Construction of Basic Education Cloud Computing Platform Based on Virtualization Technology

The information technology of basic education is constantly popularized, basic education curriculum and textbook reform are carried forward step by step, high-quality educational resources are seriously deficient, and these factors directly affect sustainable development of basic education. ‘Education cloud’ concept is introduced in the paper by transferring cloud computing technology in education field. Service frames and functions in three types of education clouds are described with virtual technique as core, in addition, concepts and methods for constructing basic education cloud platform are provided on the basis.

Minghui Zhang, Jinchen Zhou

A New Compact Teaching-Learning-Based Optimization Method

Population based heuristic optimization techniques, though powerful, are often limited by the memory size of hardware context when implemented on micro-controllers, embedded systems and commercial robots platforms etc. On the other hand, the teaching-learning based optimization algorithm (TLBO) is a recently proposed algorithm of high performance on both constrained and unconstrained optimization problems. In this paper, a new compact teachinglearning based optimization algorithm (cTLBO) is proposed to combine the strength of the original TLBO and reduce the memory requirement through a compact structure that utilizes an adaptive statistic description to replace the process of a population of solutions. Numerical results on test benchmark functions show that the new algorithm does not sacrifice the efficiency within the limited hardware resources.

Zhile Yang, Kang Li, Yuanjun Guo

ANFIS Modeling of PMV Based on Hierarchical Fuzzy System

The calculation of predicted mean vote (PMV) index is complex in real time when estimates indoor thermal comfort. As a result, some suitable model had been built to tackle this problem. In this paper, sensitivity analysis is used to sort the importance of each potential input variable on PMV. According to the results of ranking, the dimensional reduction and distribution of input space will be available. Then a T-S type hierarchical fuzzy system will be utilized to reflect PMV index by combining expert knowledge and the association analysis methods. After that the ANFIS is used to train and adjust the parameters of each subsystem through existing dataset. Simulation results show that it not only improves the accuracy but also reduce the total number of fuzzy rules.

Yifan Luo, Ning Li, Shaoyuan Li

Static Security Risk Assessment with Branch Outage Considering the Dependencies among Input Variables

This paper presents a method of static security risk assessment for wind-integrated power system with the consideration of network configuration uncertainties and correlated parameters. A probabilistic load flow (PLF) model is firstly constructed for grid-connected induction wind power system. Modeling correlated parameters and network configuration uncertainties is then taken with Cholesky decomposition, Nataf transformation, compensation method and total probability theorem. Finally, the transmission line overload risk index and the over-limit voltage index of static security are quantified, which can be used as an indicator for power system security. The proposed method is tested on the modified IEEE 30-bus system.

Xue Li, Xiong Zhang, Dajun Du

On Spacecraft Relative Orbital Motion Based on Main-Flying Direction Method

This paper introduces an analysis of a relative orbit control problem that how to let the tracking spacecraft be in the range of a small angle which is outward from the non-cooperative target spacecraft’s field of view, and at the same time satisfy the requirement of flight time and distance. This analysis is called “main-flying direction” analysis method. Based on this method, we provide a design method of the relative orbit in the fluttering mode that the tracking spacecraft flies in its own orbit after it enters into the spatial range relevant to the target spacecraft. This design method of solving the relative orbit control problem is demonstrated by simulations. The analysis thought and the calculation of this method are simple, and it avoids some disadvantages in other methods, such as the direction limitation of the field of view, attitude and orbit control coupling, etc.

Yanchao Sun, Huixiang Ling, Chuanjiang Li, Guangfu Ma, Wenrui Zhao

Special Session on Complex Networks and Their Applications

Community Detection Method of Complex Network Based on ACO Pheromone of TSP

Community detection method of complex network with a combination of TSP model and ant colony optimization is proposed in this paper. The topology relationship of network node is transformed into distance, thus the community detection problem is transformed into a path optimization problem (TSP) and solved by using ant colony algorithm, and then the pheromone matrix is used to achieve the community clustering by the convergence of algorithm. Experimental results show that, the use of TSP path length as fitness is feasible, and compared with some representative algorithms, TSPP algorithm can cluster out the number of real communities in network effectively, which has a higher clustering accuracy.

Si Liu, Cong Feng, Ming-Sheng Hu, Zhi-Juan Jia

Complex Network Construction Method of Disaster Regional Association Based on Optimized Compressive Sensing

Iming at the disaster regional association issues, a complex network construction method of disaster regional association based on compressive sensing is proposed in this paper. The disaster system dynamic equations of network node are obtained through the use of power series expansion and the correlation coefficients between nodes are obtained through the use of compressed sensing theory, then the solving process is optimized by hyperbolic tangent function and revised Newton method, so as to realize the effective construction of the network topology. Experimental results show that, complete network construction requires less amount of time series information and the construction result has a certain rationality.

Si Liu, Cong Feng, Zhi-Juan Jia, Ming-Sheng Hu

Cascading Failures in Power Grid under Three Node Attack Strategies

This paper studies cascading failures of power grid under three node attack strategies based on the local preferential redistribution rule of the broken node’s load. The initial load of a node with degree

k

is

k

β

, and

β

is a tunable parameter. We investigated the cascading propagation of US power grid under three node attack strategies and analyzed their attack effects. The three node attack strategies are HL (attack the node with the highest load), HPC (attack the node with the highest proportion between the capacity of the attacked node and the total capacities of the neighboring nodes), LL (attack the node with the lowest load). Study shows that the attack effect of HPC is the best one of the three in most cases. The attack effects of the three node attack strategies are compared with those of the three edge attack strategies. It is found that in a big range of

β

, the effect based on a node attack strategy is better than that based on the corresponding edge attack strategy. So in most cases, attacking the important nodes is more harmful than attacking the important edges in power grid.

Sui-Min Jia, Yun-Ye Wang, Cong Feng, Zhi-Juan Jia, Ming-Sheng Hu

Dynamical Distribution of Capacities Strategy for Suppressing Cascading Failure in Power Grid

This paper studies the suppressing effect of Dynamical distribution of capacities (DDC) strategy in cascading failure of power grid. This strategy is proposed based on load characteristics and transmission method of power grid. Through research of suppressing effects under 3 different mode attack modes, it is found that this strategy achieves good effect under different attacking modes and in grid with different load, but suppressing characteristics are not the same. For the strategy of attacking the node with the highest load (HL), the suppressing effect is more obvious in high load power grid while for the strategy of attacking the node with the lowest load (LL), the suppressing effect is more obvious in low load power grid. For the strategy of attacking the node with the highest proportion between the capacity of the attacked node and the total capacities of the neighboring nodes (HPC), the suppressing effect is rather obvious in any kind of load conditions in power grid.

Zhi-Juan Jia, Yu Zhang, Cong Feng, Ming-Sheng Hu

Pinning Control of Asymmetrically Coupled Complex Dynamical Network with Heterogeneous Delays

In this paper, pinning control of a class of asymmetrically coupled complex dynamical network with heterogeneous delays is studied. Two kinds of controllers, adaptive controllers and linear feedback controllers, are presented and the Jordan canonical transformation method is used instead of the matrix diagonalization method. Therefore, it is not necessary for some relevant matrices to be diagonalizable. Moreover, a simply approximate formula is provided to estimate how many and which nodes should a network with fixed network structure and coupling strength be pinned to reach synchronization. Here, the inner-coupling matrix is not necessarily symmetric. One example is given to show the effectiveness of the proposed synchronization criteria.

Fengli Ren, Hongyong Zhao

Particle Swarm Optimizations for Multi-type Vehicle Routing Problem with Time Windows

This paper presents a variant of vehicle routing problem with time windows (VRPTW) named multi-type vehicle routing problem with time windows (MT-VRPTW), which considers both multiple types of the vehicle and the uncertain number of vehicles of various types. As a consequence, the different combinations of multi-type vehicle will lead to diverse results, which should be evaluated by its own fitness function. In order to solve the proposed MT-VRPTW problem, six variants of particle swarm optimization (PSO) are used. The 2N dimensions encoding method is adopted to express the particle (N represents the number of demand point). In the simulation studies, the performances of the six PSO variants are compared and the obtained results are analyzed.

Xiaobin Gan, Junbiao Kuang, Ben Niu

Special Session on Time Series Forecasting and Analysis Using Artificial Neural Networks

Prediction of Physical Time Series Using Spiking Neural Networks

Forecasting the behavior of naturally occurring phenomena by the analysis of time series based data is the basis of scientific experimental design. In this paper, we consider a novel application of a Polychronous Spiking Network for the prediction of sunspot and auroral electrojet index by exploiting the inherent temporal capabilities of this spiking neural model. The performance of this network is benchmarked against two “traditional”, rateencoded, neural networks; a Multi-Layer Perceptron network and a Functional Link Neural Network. The results indicate that the inherent temporal characteristics of the Polychronous Spiking Network make it extremely well suited to the processing of time series based data.

David Reid, Abir Jaafar Hussain, Hissam Tawfik, Rozaida Ghazali

The Application of Artificial Immune Systems for the Prediction of Premature Delivery

One of the most challenging tasks currently facing the healthcare community is the identification of premature labour. Premature birth occurs when the baby is born before completion of the 37-week gestation period. The incomplete understanding of the physiology of the uterus and parturition means that premature labour prediction is a difficult task. The reason for this may be that the initial symptoms of preterm labour occur commonly in normal pregnancies. There is some misclassification in regard to recognizing full-term and preterm labour; approximately 20% of women who are identified as reaching full-term labour actually deliver prematurely. This paper explores the applicability of Artificial Immune System (AIS) technique as a new methodology to classify term and preterm records. Our AIS approach shows better results when compared with Neural Network, Decision Tree, and Support Vector Machines, achieving more than 92% accuracy overall.

Rentian Huang, Hissam Tawfik, Abir Jaafar Hussain, Haya Al-Askar

Special Session on Computer Human Interaction Using Multiple Visual Cues and Intelligent Computing

Dynamic Hand Gesture Recognition Framework

Sign languages originated long before any speech-based languages evolved in the world. They contain subtleties that rival any speech-based languages conveying a rich source of information much faster than any speechbased languages. Similar to the diversity of speech-based languages, sign languages vary from region to region. However, unlike the speech counterpart, sign languages from diverse regions from the world have much common traits that originate from human evolution. Researchers have been intrigued by these common traits and have always wondered whether sign language-type communication is possible for instructing the computers opposed to the mundane keyboard and mouse. This trend is popularly known as Human Computer Interaction (HCI) and has used a subset of common sign language hand gestures to interact with machines through computer vision.Since the sign languages comprise of thousands of subtle gestures, a new sophisticated approach has to be initiated for eventual recognition of vast number of gestures. Hand gestures comprise of both static postures and dynamic gestures and can carry significantly rich vocabulary describing words in the thousands. In this article, we present our latest research that describes a mechanism to accurately interpret dynamic hand gestures using a concept known as ‘gesture primitives’ where each dynamic gesture is described as a collection of many primitives over time that can drive a classification strategy based on Hidden Markov Model to reliably predict the gesture using statistical knowledge of such gestures. We believe that even though our work is in its infancy, this strategy can be extended to thousands of dynamic gestures used in sign language to be interpreted by machines.

Prashan Premaratne, Shuai Yang, ZhengMao Zhou, Nalin Bandara

Special Session on Biometric System and Security for Intelligent Computing

No-Reference Fingerprint Image Quality Assessment

Quality of a fingerprint image is assessed to control the registration of poor quality images in the database so that a good accuracy of fingerprint recognition system can be achieved. This paper proposes a quality assessment scheme for digital fingerprint image. It makes use of complete ridge line of a thinned fingerprint image for quality assessment. It introduces three robust measures (1) ridge-line smoothness (2) inter ridge-line distance and (3) minutiae extractability for the quality assessment. Experiments are performed on a database comprising of 1000 fingerprint images of 500 subjects of various age group lying between 18 and 75. It has been found that the performance of the fingerprint recognition system is improved from CRR of 74.6% to 100% and EER of 13.08% to 0.01% by controlling the registration of inferior quality fingerprint in the database. Quality of a fingerprint images is an important indicator of its performance in automatic fingerprint based recognition system.

Kamlesh Tiwari, Phalguni Gupta

Backmatter

Weitere Informationen