Skip to main content
main-content

Über dieses Buch

The three volume set LNCS 8834, LNCS 8835, and LNCS 8836 constitutes the proceedings of the 20th International Conference on Neural Information Processing, ICONIP 2014, held in Kuching, Malaysia, in November 2014. The 231 full papers presented were carefully reviewed and selected from 375 submissions. The selected papers cover major topics of theoretical research, empirical study, and applications of neural information processing research. The 3 volumes represent topical sections containing articles on cognitive science, neural networks and learning systems, theory and design, applications, kernel and statistical methods, evolutionary computation and hybrid intelligent systems, signal and image processing, and special sessions intelligent systems for supporting decision, making processes,theories and applications, cognitive robotics, and learning systems for social network and web mining.

Inhaltsverzeichnis

Frontmatter

Kernel and Statistical Methods

A New Ensemble Clustering Method Based on Dempster-Shafer Evidence Theory and Gaussian Mixture Modeling

This paper proposes a new method based on Dempster-Shafer (DS) evidence theory and Gaussian Mixture Modeling (GMM) technique to combine the cluster results from single clustering methods. We introduce the GMM technique to determine the confidence values for candidate results from each clustering method. Then we employ the DS theory to combine the evidences supplied by different clustering methods, based on which the final result is obtained. We tested the proposed ensemble clustering method on several commonly used datasets. The experimental results confirm that our method is effective and promising.

Yi Wu, Xiabi Liu, Lunhao Guo

Extraction of Dimension Reduced Features from Empirical Kernel Vector

This paper proposes a feature extraction method from the given empirical kernel vector. We show the necessary condition for the feature extraction mapping to make the trained classifier by using the linear SVM with the extracted feature vectors equivalent to the one obtained by the standard kernel SVM. The proposed feature extraction mapping is defined by using the eigen values and eigen vectors of the Gram matrix. Since the eigen vector problem of the Gram matrix is closely related with the kernel Principal Component Analysis, we can extract a dimension reduced feature vector. This feature extraction method becomes equivalent to the kernel SVM if the full dimension is used. The proposed feature extraction method was evaluated by the experiments using the standard data sets. The cross-validation values of the proposed method were improved and the recognition rates were comparable with the original kernel SVM. The number of extracted features was very low compared to the number of features of the kernel SVM.

Takio Kurita, Yayoi Harashima

Method of Evolving Non-stationary Multiple Kernel Learning

Recently, evolving multiple kernel learning methods have attracted researchers’ attention due to the ability to find the composite kernel with the optimal mapping model in a large high-dimensional feature space. However, it is not suitable to compute the composite kernel in a stationary way for all samples. In this paper, we propose a method of evolving non-stationary multiple kernel learning, in which base kernels are encoded as tree kernels and a gating function is used to determine the weights of the tree kernels simultaneously. Obtained classifiers have the composite kernel with the optimal mapping model and select the most appropriate combined weights according to the input samples. Experimental results on several UCI datasets illustrate the validity of proposed method.

Peng Wu, Qian Yin, Ping Guo

A Kernel Method to Extract Common Features Based on Mutual Information

Kernel canonical correlation analysis (CCA) aims to extract common features from a pair of multivariate data sets by maximizing a linear correlation between nonlinear mappings of the data. However, the kernel CCA tends to obtain the features that have only small information of original multivariates in spite of their high correlation, because it considers only statistics of the extracted features and the nonlinear mappings have high degree of freedom. We propose a kernel method for common feature extraction based on mutual information that maximizes a new objective function. The objective function is a linear combination of two kinds of mutual information, one between the extracted features and the other between the multivariate and its feature. A large value of the former mutual information provides strong dependency to the features, and the latter prevents loss of the feature’s information related to the multivariate. We maximize the objective function by using the Parallel Tempering MCMC in order to overcome a local maximum problem. We show the effectiveness of the proposed method via numerical experiments.

Takamitsu Araki, Hideitsu Hino, Shotaro Akaho

Properties of Text-Prompted Multistep Speaker Verification Using Gibbs-Distribution-Based Extended Bayesian Inference for Rejecting Unregistered Speakers

This paper presents a method of text-prompted multistep speaker verification for reducing verification errors and rejecting unregistered speakers. The method has been developed for our speech processing system which utilizes competitive associative nets (CAN2s) for learning piecewise linear approximation of nonlinear speech signal to extract feature vectors of pole distribution from piecewise linear coefficients reflecting nonlinear and time-varying vocal tract of the speaker. This paper focuses on rejecting unregistered speakers by means of multistep verification using Gibbs-distribution-based extended Bayesian inference (GEBI) in text-prompted speaker verification. The properties of GEBI and the comparison to BI (Bayesian inference) for rejecting unregistered speakers are shown and analyzed by means of experiments using real speech signals.

Shuichi Kurogi, Takuya Ueki, Satoshi Takeguchi, Yuta Mizobe

Non-monotonic Feature Selection for Regression

Feature selection is an important research problem in machine learning and data mining. It is usually constrained by the budget of the feature subset size in practical applications. When the budget changes, the ranks of features in the selected feature subsets may also change due to nonlinear cost functions for acquisition of features. This property is called non-monotonic feature selection. In this paper, we focus on non-monotonic selection of features for regression tasks and approximate the original combinatorial optimization problem by a Multiple Kernel Learning (MKL) problem and show the performance guarantee for the derived solution when compared to the global optimal solution for the combinatorial optimization problem. We conduct detailed experiments to demonstrate the effectiveness of the proposed method. The empirical results indicate the promising performance of the proposed framework compared with several state-of-the-art approaches for feature selection.

Haiqin Yang, Zenglin Xu, Irwin King, Michael R. Lyu

Non-negative Matrix Factorization with Schatten p-norms Reguralization

In this paper we study the effect of regularization on clustering results provided by Non-negative Matrix Factorization (NMF). Different kinds of regularization terms were previously added to the NMF objective function in order to produce sparser results and thus to obtain a more qualitative partition of data. We would like to propose the general framework for regularized NMF based on Schatten p-norms. Experimental results show the effectiveness of our approach on different data sets.

Ievgen Redko, Younès Bennani

A New Energy Model for the Hidden Markov Random Fields

In this article we propose a modification to the HMRF-EM framework applied to image segmentation. To do so, we introduce a new model for the neighborhood energy function of the Hidden Markov Random Fields model based on the Hidden Markov Model formalism. With this new energy model, we aim at (1) avoiding the use of a key parameter chosen empirically on which the results of the current models are heavily relying, (2) proposing an information rich modelisation of neighborhood relationships.

Jérémie Sublime, Antoine Cornuéjols, Younès Bennani

Online Nonlinear Granger Causality Detection by Quantized Kernel Least Mean Square

Identifying causal relations among simultaneously acquired signals is an important challenging task in time series analysis. The original definition of Granger causality was based on linear models, its application to nonlinear systems may not be appropriate. We consider an extension of Granger causality to nonlinear bivariate time series with the universal approximation capacity in reproducing kernel Hilbert space (RKHS) while preserving the conceptual simplicity of the linear model. In particular, we propose a computationally simple online measure by means of quantized kernel least mean square (QKLMS) to capture instantaneous causal relationships.

Hong Ji, Badong Chen, Zejian Yuan, Nanning Zheng, Andreas Keil, Jose C. Príncipe

A Computational Model of Anti-Bayesian Sensory Integration in the Size-Weight Illusion

We propose a computational model for anti-Bayesian sensory integration of human behavioral actions and perception in the size–weight illusion (SWI). The SWI refers to the fact that people judge the smaller of two equally weighted objects to heavier when lifted. Many aspects of human perceptual and motor behavior can be modeled with Bayesian statistics. However, the SWI cannot be explained on the basis of Bayesian integration, and the nervous system is thought to use two entirely different mechanisms to integrate prior expectations with current sensory information about object weight. Our proposed model is defined as a state estimator, combining a Kalman filter and a H

 ∞ 

filter. As a result, the model not only predicted the anti-Bayesian estimation of the weight but also the Bayesian estimation of the motor behavior. Therefore, we hypothesize that the SWI is realized by a H

 ∞ 

filter and a Kalman filter.

Yuki Ueyama

Unsupervised Dimensionality Reduction for Gaussian Mixture Model

Dimensionality reduction is a fundamental yet active research topic in pattern recognition and machine learning. On the other hand, Gaussian Mixture Model (GMM), a famous model, has been widely used in various applications, e.g., clustering and classification. For high-dimensional data, previous research usually performs dimensionality reduction first, and then inputs the reduced features to other available models, e.g., GMM. In particular, there are very few investigations or discussions on how dimensionality reduction could be interactively and systematically conducted together with the important GMM. In this paper, we study the problem how unsupervised dimensionality reduction could be performed together with GMM and if such joint learning could lead to improvement in comparison with the traditional unsupervised method. Specifically, we engage the Mixture of Factor Analyzers with the assumption that a common factor loading exist for all the components. Such setting exactly optimizes a dimensionality reduction together with the parameters of GMM. We compare the joint learning approach and the separate dimensionality reduction plus GMM method on both synthetic data and real data sets. Experimental results show that the joint learning significantly outperforms the comparison method in terms of three criteria for supervised learning.

Xi Yang, Kaizhu Huang, Rui Zhang

Graph Kernels Exploiting Weisfeiler-Lehman Graph Isomorphism Test Extensions

In this paper we present a novel graph kernel framework inspired the by the Weisfeiler-Lehman (WL) isomorphism tests. Any WL test comprises a relabelling phase of the nodes based on test-specific information extracted from the graph, for example the set of neighbours of a node. We defined a novel relabelling and derived two kernels of the framework from it. The novel kernels are very fast to compute and achieve state-of-the-art results on five real-world datasets.

Giovanni Da San Martino, Nicolò Navarin, Alessandro Sperduti

Texture Analysis Based Automated Decision Support System for Classification of Skin Cancer Using SA-SVM

Early diagnosis of skin cancer is one of the greatest challenges due to lack of experience of general practitioners (GPs). This paper presents a clinical decision support system aimed to save lives, time and resources in the early diagnostic process. Segmentation, feature extraction, and lesion classification are the important steps in the proposed system. The system analyses the images to extract the affected area using a novel proposed segmentation method H-FCM-LS. A set of 45 texture based features is used. These underlying features which indicate the difference between melanoma and benign images are obtained through specialized texture analysis methods. For classification purpose, self-advising SVM is adapted which showed improved classification rate as compared to standard SVM. The diagnostic accuracy obtained through the proposed system is around 90% with sensitivity 91% and specificity 89%.

Ammara Masood, Adel Al-Jumaily, Khairul Anam

In-attention State Monitoring for a Driver Based on Head Pose and Eye Blinking Detection Using One Class Support Vector Machine

This paper proposes a model to detect inattention cognitive state of a driver during various driving situations. The proposed system predicts driver’s inattention state based on the analysis of eye blinking patterns and head pose direction. The study uses an infrared camera and several feature extraction stages such as modified census transform (MCT) to reduce the effect of light source change in real traffic environment. Also, we propose a new eye blinking detection using the difference between center and surround of Hough circle transform image. The local linear embedding (LLE) is used to extract real-time features of head movement. Finally, the driver’s cognitive states can be estimated by the one-class support vector machines (OCSVMs) using both eyes blinking patterns and head pose direction information. We implement a prototype of the proposed driver state monitoring (DSM) system. Experimental results show that the proposed system using OCSVM works well in real environment compared to the system that employs SVM.

Hyunrae Jo, Minho Lee

An Improved Separating Hyperplane Method with Application to Embedded Intelligent Devices

Classification is a common task in pattern recognition. Classifiers used in embedded intelligent devices need a good trade-off between prediction accuracy, resource consumption and prediction speed. Support vector machine(SVM) is accurate but its run-time complexity is higher due to the large number of support vectors. A new separating hyperplane method (NSHM) for the binary classification task was proposed. NSHM allows fast classification. However, NSHM is order-sensitive and this affects its classification accuracy. Inspired by NSHM, we propose CSHM, a combining separating hyperplane method. CSHM combines all optimal separating hyperplanes found by NSHM. Experimental results on UCI Machine Learning Repository show that, compared with NSHM and SVM, CSHM achieves a better trade-off between prediction accuracy, resource consumption and prediction speed.

Yanjun Li, Ping Guo, Xin Xin

Fine-Grained Air Quality Monitoring Based on Gaussian Process Regression

Air quality is attracting more and more attentions in recent years due to the deteriorating environment, and

PM

2.5

is the main contaminant in a lot of areas. Existing softwares that report the level of

PM

2.5

can provide only the value in the city level, which may indeed varies greatly among different areas in the city. To help people know about the exact air quality around them, we deployed 51 carefully designed devices to measure the

PM

2.5

at these places and present a Gaussian Process based inference model to estimate the value at any place. The proposed method is evaluated on the real data and compared to some related methods. The experimental results prove the effectiveness of our method.

Yun Cheng, Xiucheng Li, Zhijun Li, Shouxu Jiang, Xiaofan Jiang

Retrieval of Experiments by Efficient Comparison of Marginal Likelihoods

We study the task of retrieving relevant experiments given a query experiment. By experiment, we mean a collection of measurements from a set of ‘covariates’ and the associated ‘outcomes’. While similar experiments can be retrieved by comparing available ‘annotations’, this approach ignores the valuable information available in the measurements themselves. To incorporate this information in the retrieval task, we suggest employing a retrieval metric that utilizes probabilistic models learned from the measurements. We argue that such a metric is a sensible measure of similarity between two experiments since it permits inclusion of experiment-specific prior knowledge. However, accurate models are often not analytical, and one must resort to storing posterior samples which demands considerable resources. Therefore, we study strategies to select informative posterior samples to reduce the computational load while maintaining the retrieval performance. We demonstrate the efficacy of our approach on simulated data with simple linear regression as the models, and real world datasets.

Sohan Seth, John Shawe-Taylor, Samuel Kaski

Evolutionary Computation and Hybrid Intelligent Systems

A New Approach of Diversity Enhanced Particle Swarm Optimization with Neighborhood Search and Adaptive Mutation

Like other stochastic algorithms, particle swarm optimization algorithm (PSO) has shown a good performance over global numerical optimization. However, PSO also has a few drawbacks such as premature convergence and low convergence speed, especially on complex problem. In this paper, we present a new approach called AMPSONS in which neighborhood search, diversity mechanism and adaptive mutation were utilized. Experimental results obtained from a test on several benchmark functions showed that the performance of proposed AMPSONS algorithm is superior to five other PSO variants, namely CLPSO, AMPSO, GOPSO, DNLPSO, and DNSPSO, in terms of convergence speed and accuracy.

Dang Cong Tran, Zhijian Wu, Hui Wang

Data Clustering Based on Particle Swarm Optimization with Neighborhood Search and Cauchy Mutation

K-means is one of the most popular clustering algorithm, it has been successfully applied in solving many practical clustering problems, however there exist some drawbacks such as local optimal convergence and sensitivity to initial points. In this paper, a new approach based on enhanced particle swarm optimization (PSO) is presented (denoted CMPNS), in which PSO is enhanced by new neighborhood search strategy and Cauchy mutation operation. Experimental results on fourteen used artificial and real-world datasets show that the proposed method outperforms than that of some other data clustering algorithms in terms of accuracy and convergence speed.

Dang Cong Tran, Zhijian Wu

Accuracy Improvement of Localization and Mapping of ICP-SLAM via Competitive Associative Nets and Leave-One-Out Cross-Validation

This paper presents a method to improve the accuracy of localization and mapping obtained by ICP-SLAM (iterative closest point - simultaneous localization and mapping) algorithm. The method uses competitive associative net (CAN2) for learning piecewise linear approximation of the cloud of 2D points obtained by the LRF (laser range finder) mounted on a mobile robot. To reduce the propagation error caused by the consecutive pairwise registration by the ICP-SLAM algorithm, the present method utilizes leave-one-out cross-validation (LOOCV) and tries to minimize the LOOCV registration error. The effectiveness is shown by analyzing the real experimental data.

Shuichi Kurogi, Yoichiro Yamashita, Hikaru Yoshikawa, Kotaro Hirayama

Saliency Level Set Evolution

In this paper, we consider saliency detection problems from a unique perspective. We provide an implicit representation for the saliency map using level set evolution (LSE), and then combine LSE approach with energy functional minimization (EFM). Instead of introducing sophisticated segmentation procedures, we propose a flexible and lightweight LSE-EFM framework for saliency detection. The experimental results demonstrate our method outperforms several existing popular approaches. We then evaluate several computation strategies independently. The comparisons results indicate their effectiveness and strong abilities in combatting saliency confusions.

Jincheng Mei, Bao-Liang Lu

Application of Cuckoo Search for Design Optimization of Heat Exchangers

A wide variety of evolutionary optimization algorithms have been used by researcher for optimal design of shell and tube heat exchangers (STHX). The purpose of optimization is to minimize capital and operational costs subject to efficiency constraints. This paper comprehensively examines performance of genetic algorithm (GA) and cuckoo search (CS) for solving STHX design optimization. While GA has been widely adopted in the last decade for STHX optimal design, there is no report on application of CS method for this purpose. Simulation results in this paper demonstrate that CS greatly outperforms GA in terms of finding admissible and optimal configurations for STHX. It is also found that CS method not only has a lower computational requirement, but also generates the most consistent results.

Rihanna Khosravi, Abbas Khosravi, Saeid Nahavandi

A Hybrid Method to Improve the Reduction of Ballistocardiogram Artifact from EEG Data

Simultaneous recordings of functional magnetic resonance imaging (fMRI) and electroencephalography (EEG) allow acquisition of brain data with high spatial and temporal resolution. However, the EEG data get contaminated by additional artifacts such as Gradient artifact and Ballistocardiogram (BCG) artifact. The BCG artifact’s dynamics appear to be more challenging and it hinders in the assessment of the neuronal activities. In this paper, a reference-free method is implemented in which Empirical Mode Decomposition (EMD) and Principal Component Analysis (PCA) has been combined to reduce the BCG artifact while preserving the neuronal activities. The qualitative analysis of the proposed method along with three existing methods demonstrates that the proposed method has improved the quality of the reconstructed data. Moreover, it does not require any reference signal to extract BCG artifact.

Ehtasham Javed, Ibrahima Faye, Aamir Saeed Malik, Jafri Malin Abdullah

VLGAAC: Variable Length Genetic Algorithm Based Alternative Clustering

Complex and heterogeneous data sets can often be interpreted as having multiple clustering, each of which are valid but distinct from the others. Several algorithms involving multiple objective functions have been reported for such alternative clusterings. We propose a genetic algorithm based approach for obtaining valid but diverse clustering. A variable length genetic algorithm approach is used to enable varying number of clusters in each interpretation. A suitable method for population initialization and appropriate crossover and mutation operators are also used. Experimental results on benchmark data sets show that the method is comparable with related alternative clustering techniques.

Moumita Saha, Pabitra Mitra

Social Book Search with Pseudo-Relevance Feedback

Massive books with social information, e.g. reviews, rates and tags, have emerged in large numbers on the web. However, there are several limitations in traditional search methods for social books, as social books include complicated and various social information. Relevance feedback is always an important and concerned technique in information retrieval. Therefore in this paper we propose a search system based on pseudo-relevance feedback (PRF) for expanding and enriching the social information of queries. In our system, First, Galago is used to get the initial rank list. Then relevance models are performed to select candidate high-frequent words that can be benefit to queries. Next, the original queries and these selected words are combined into new queries by linear smoothing. With evaluation on the INEX2012 / 2013 Social Book Search Track database, our proposed system has an encouraged performance (nDCG@10) compared to several state-of-the-art (contest) systems.

Bin Geng, Fang Zhou, Jiao Qu, Bo-Wen Zhang, Xiao-Ping Cui, Xu-Cheng Yin

A Random Key Genetic Algorithm for Live Migration of Multiple Virtual Machines in Data Centers

Live migration of multiple Virtual Machines (VMs) has become an integral management activity in data centers for power saving, load balancing and system maintenance. While state-of-the-art live migration techniques focus on the improvement of migration performance of an independent single VM, only a little has been investigated to the case of live migration of multiple interacting VMs. Live migration is mostly influenced by the network bandwidth and arbitrarily migrating a VM which has data inter-dependencies with other VMs may increase the bandwidth consumption and adversely affect the performances of subsequent migrations. In this paper, we propose a Random Key Genetic Algorithm (RKGA) that efficiently schedules the migration of a given set of VMs accounting both inter-VM dependency and data center communication network. The experimental results show that the RKGA can schedule the migration of multiple VMs with significantly shorter total migration time and total downtime compared to a heuristic algorithm.

Tusher Kumer Sarker, Maolin Tang

Collaboration of the Radial Basis ART and PSO in Multi-Solution Problems of the Hénon Map

This paper studies collaboration of the ART and PSO in application to a multi-solution problem for analysis of the H

$\acute{\mbox{e}}$

non map. In our algorithm, the PSO gives candidates of solutions which have no labels. Applying the candidates as inputs, the ART classifies the candidates, labels the categories, and clarify the number of solutions. Performing fundamental numerical experiments, the algorithm efficiency is investigated.

Fumiaki Tokunaga, Takumi Sato, Toshimichi Saito

Reconstructing Gene Regulatory Network with Enhanced Particle Swarm Optimization

Inferring regulations among the genes is a well-known and significantly important problem in systems biology for revealing the fundamental cellular processes. Although computational models can be used as tools to extract the probable structure and dynamics of such networks from gene expression data, capturing the complex nonlinear system dynamics is a challenging task. In this paper, we have proposed a method to reverse engineering Gene Regulatory Network (GRN) from microarray data. Inspired from the biologically relevant optimization algorithm ‘Particle Swarm Optimization’ (PSO), we have enhanced the PSO incorporating two genetic algorithm operators, namely crossover and mutation. Furthermore, Linear Time Variant (LTV) Model is employed to modeling the GRN appropriately. In the evaluation, the proposed method shows superiority over the state-of-the-art methods when tested with synthetic network, both for the noise free and noise in data. The strength of the proposed method has also been verified by analyzing the real expression data set of SOS DNA repair system in

Escherichia coli

.

Rezwana Sultana, Dilruba Showkat, Mohammad Samiullah, Ahsan Raja Chowdhury

Neural Network Training by Hybrid Accelerated Cuckoo Particle Swarm Optimization Algorithm

Metaheuristic algorithm is one of the most popular methods in solving many optimization problems. This paper presents a new hybrid approach comprising of two natures inspired metaheuristic algorithms i.e. Cuckoo Search (CS) and Accelerated Particle Swarm Optimization (APSO) for training Artificial Neural Networks (ANN). In order to increase the probability of the egg’s survival, the cuckoo bird migrates by traversing more search space. It can successfully search better solutions by performing levy flight with APSO. In the proposed Hybrid Accelerated Cuckoo Particle Swarm Optimization (HACPSO) algorithm, the communication ability for the cuckoo birds have been provided by APSO, thus making cuckoo bird capable of searching for the best nest with better solution. Experimental results are carried-out on benchmarked datasets, and the performance of the proposed hybrid algorithm is compared with Artificial Bee Colony (ABC) and similar hybrid variants. The results show that the proposed HACPSO algorithm performs better than other algorithms in terms of convergence and accuracy.

Nazri Mohd Nawi, Abdullah khan, M. Z. Rehman, Maslina Abdul Aziz, Tutut Herawan, Jemal H. Abawajy

An Accelerated Particle Swarm Optimization Based Levenberg Marquardt Back Propagation Algorithm

The Levenberg Marquardt (LM) algorithm is one of the most effective algorithms in speeding up the convergence rate of the Artificial Neural Networks (ANN) with Multilayer Perceptron (MLP) architectures. However, the LM algorithm suffers the problem of local minimum entrapment. Therefore, we introduce several improvements to the Levenberg Marquardt algorithm by training the ANNs with meta-heuristic nature inspired algorithm. This paper proposes a hybrid technique Accelerated Particle Swarm Optimization using Levenberg Marquardt (APSO_LM) to achieve faster convergence rate and to avoid local minima problem. These techniques are chosen since they provide faster training for solving pattern recognition problems using the numerical optimization technique.The performances of the proposed algorithm is evaluated using some bench mark of classification’s datasets. The results are compared with Artificial Bee Colony (ABC) Algorithm using Back Propagation Neural Network (BPNN) algorithm and other hybrid variants.Based on the experimental result, the proposed algorithms APSO_LM successfully demonstrated better performance as compared to other existing algorithms in terms of convergence speed and Mean Squared Error (MSE) by introducing the error and accuracy in network convergence.

Nazri Mohd Nawi, Abdullah khan, M. Z. Rehman, Maslina Abdul Aziz, Tutut Herawan, Jemal H. Abawajy

Fission-and-Recombination Particle Swarm Optimizers for Search of Multiple Solutions

This paper presents the fission-and-recombination particle swarm optimizer (FRPSO) and its application to search of periodic points of a nonlinear dynamical system. The search problem is translated into a multi-solution problem evaluated in a multi-objective problem. The FRPSO is based on the ring topology. The FRPSO is effective to escape from trap of partial/local solutions and to find all the solutions. Performing basic numerical experiments, the algorithm efficiency is investigated.

Takumi Sato, Toshimichi Saito

Fast Generalized Fuzzy C-means Using Particle Swarm Optimization for Image Segmentation

Fuzzy C-means algorithms (FCMs) incorporating local information has been widely used for image segmentation, especially on image corrupted by noise. However, they cannot obtain the satisfying segmentation performance on the image heavily contaminated by noise, sensitivity to initial points, and can be trapped into local optima. Hence, optimization techniques are often used in conjunction with algorithms to improve the performance. In this paper, Particle Swarm Optimization (PSO) is introduced into fast generalized FCM (FGFCM) incorporating with local spatial and gray information called PFGFCM, where the membership degree values were modified by applying optimal-selection-based suppressed strategy. Experimental results on synthetic and real images heavily corrupted by noise show that the proposed method is superior to other fuzzy algorithms.

Dang Cong Tran, Zhijian Wu, Van Hung Tran

Evolutionary Learning and Stability of Mixed-Rule Cellular Automata

This paper studies the cellular automaton (CA) governed by combination of two rules. First, we analyze a class of CA that generates several isolated spatiotemporal patterns without transient phenomena. Second, we present an evolutionary algorithm that tries to optimize the combination of two rules to stabilize the desired isolated patterns. Performing basic numerical experiments, it is shown that the evolutionary algorithm can make transient phenomena and can stabilize the desired isolated patterns.

Ryo Sawayama, Toshimichi Saito

Pattern Recognition Techniques

Radical-Enhanced Chinese Character Embedding

In this paper, we present a method to leverage radical for learning Chinese character embedding. Radical is a semantic and phonetic component of Chinese character. It plays an important role for modelling character semantics as characters with the same radical usually have similar semantic meaning and grammatical usage. However, most existing character (or word) embedding learning algorithms typically only model the syntactic contexts but ignore the radical information. As a result, they do not explicitly capture the inner semantic connections of characters via radical into the embedding space of characters. To solve this problem, we propose to incorporate the radical information for enhancing the Chinese character embedding. We present a dedicated neural architecture with a hybrid loss function, and integrate the radical information through softmax upon each character. To verify the effectiveness of the learned character embedding, we apply it on Chinese word segmentation. Experiment results on two benchmark datasets show that, our radical-enhanced method outperforms two widely-used context-based embedding learning algorithms.

Yaming Sun, Lei Lin, Nan Yang, Zhenzhou Ji, Xiaolong Wang

Conditional Multidimensional Parameter Identification with Asymmetric Correlated Losses of Estimation Errors

This paper is dedicated to the problem of the estimation of a vector of parameters, as losses resulting from their under- and overestimation are asymmetric and mutually correlated. The issue is considered from an additional conditional aspect, where particular coordinates of conditioning variables may be continuous, binary, discrete or categorized (ordered and unordered). The final result is an algorithm for calculating the value of an estimator, optimal in sense of expectation of losses using a multidimensional asymmetric quadratic function, for practically any distributions of describing and conditioning variables.

Piotr Kulczycki, Malgorzata Charytanowicz

Short Text Hashing Improved by Integrating Topic Features and Tags

Hashing, as an efficient approach, has been widely used for large-scale similarity search. Unfortunately, many existing hashing methods based on observed keyword features are not effective for short texts due to the sparseness and shortness. Recently, some researchers try to construct semantic relationship using certain granularity topics. However, the topics of certain granularity are insufficient to preserve the optimal semantic similarity for different types of datasets. On the other hand, tag information should be fully exploited to enhance the similarity of related texts. We, therefore, propose a novel unified hashing approach that the optimal topic features can be selected automatically to be integrated with original features for preserving similarity, and tags are fully utilized to improve hash code learning. We carried out extensive experiments on one short text dataset and even one normal text dataset. The results demonstrate that our approach is effective and significantly outperforms baseline methods on several evaluation metrics.

Jiaming Xu, Bo Xu, Jun Zhao, Guanhua Tian, Heng Zhang, Hongwei Hao

Synthetic Test Data Generation for Hierarchical Graph Clustering Methods

Recent achievements in graph-based clustering algorithms revealed the need for large-scale test data sets. This paper introduces a procedure that can provide synthetic but realistic test data to the hierarchical Markov clustering algorithm. Being created according to the structure and properties of the SCOP95 protein sequence data set, the synthetic data act as a collection of proteins organized in a four-level hierarchy and a similarity matrix containing pairwise similarity values of the proteins. An ultimate high-speed TRIBE-MCL algorithm was employed to validate the synthetic data. Generated data sets have a healthy amount of variability due to the randomness in the processing, and are suitable for testing graph-based clustering algorithms on large-scale data.

László Szilágyi, Levente Kovács, Sándor Miklós Szilágyi

Optimal Landmark Selection for Nyström Approximation

The Nyström method is an efficient technique for large-scale kernel learning. It provides a low-rank matrix approximation to the full kernel matrix. The quality of Nyström approximation largely depends on the choice of landmark points. While standard method uniformly samples columns of the kernel matrix, improved sampling techniques have been proposed based on ensemble learning [1] and clustering [2]. These methods are focused on minimizing the approximation error for the original kernel. In this paper, we take a different perspective by minimizing the approximation error for the input vectors instead. We show under some restrictive condition that the new formulation is equivalent to the standard Nyström solution. This leads to a novel approach for optimizing landmark points for the Nyström approximation. Experimental results demonstrate the superior performance of the proposed landmark optimization method compared to existing Nyström methods in terms of lower approximation errors obtained.

Zhouyu Fu

Privacy Preserving Clustering: A k-Means Type Extension

We study the problem of

r

-anonymized clustering and give a

k

-means type extension. The problem is partition a set of objects into

k

different groups by minimizing the total cost between objects and cluster centers subject to a constraint that each cluster contains at least

r

objects. Previous work has reported an approach when the cluster centers are constrained to be a real member of the objects. In this paper, we release the constraint and allow a center to be the mean of the objects in its group, similar to the settings of the classical

k

-means clustering model. To address the inherent computational difficulty, we exploit linear program relaxation to find high quality solutions in an efficient manner. We conduct a series of experiments and confirm the effectiveness of the method as expected.

Wenye Li

Stream Quantiles via Maximal Entropy Histograms

We address the problem of estimating the running quantile of a data stream when the memory for storing observations is limited. We (i) highlight the limitations of approaches previously described in the literature which make them unsuitable for non-stationary streams, (ii) describe a novel principle for the utilization of the available storage space, and (iii) introduce two novel algorithms which exploit the proposed principle. Experiments on three large real-world data sets demonstrate that the proposed methods vastly outperform the existing alternatives.

Ognjen Arandjelović, Ducson Pham, Svetha Venkatesh

A Unified Framework for Thermal Face Recognition

The reduction of the cost of infrared (IR) cameras in recent years has made IR imaging a highly viable modality for face recognition in practice. A particularly attractive advantage of IR-based over conventional, visible spectrum-based face recognition stems from its invariance to visible illumination. In this paper we argue that the main limitation of previous work on face recognition using IR lies in its

ad hoc

approach to treating different nuisance factors which affect appearance, prohibiting a unified approach that is capable of handling concurrent changes in multiple (or indeed all) major extrinsic sources of variability, which is needed in practice. We describe the first approach that attempts to achieve this – the framework we propose achieves outstanding recognition performance in the presence of variable (i) pose, (ii) facial expression, (iii) physiological state, (iv) partial occlusion due to eye-wear, and (v) quasi-occlusion due to facial hair growth.

Reza Shoja Ghiass, Ognjen Arandjelović, Hakim Bendada, Xavier Maldague

Geometric Feature-Based Facial Emotion Recognition Using Two-Stage Fuzzy Reasoning Model

Facial Emotion recognition is a significant requirement in machine vision society. In this sense, this paper utilizes geometric facial features and calculates displacement of feature points between expressive and neutral frames and finally applies a two-stage fuzzy reasoning model for facial emotion recognition and classification. The prototypical emotion sequence according to the Facial Action Coding System (FACS) is formed analyzing small, medium and large displacement. Furthermore geometric displacements are fuzzified and mapped onto an Action Units (AUs) by employing first-stage fuzzy reasoning model and later AUs are fuzzified and mapped onto an Emotion space by employing second-stage fuzzy relational model. The overall performance of the proposed system is evaluated on the extended Cohn-Kanade (CK+) database for classifying basic emotions like surprise, sadness, fear, anger, and happiness. The experimental results on the task of facial emotion analysis and emotion recognition are shown to outperform other existing methods available in the literature.

Md. Nazrul Islam, Chu Kiong Loo

Human Activity Recognition by Matching Curve Shapes

In this paper, we present a new method for Human Activity Recognition (HAR) from body-worn accelerometers or inertial sensors using comparison of curve shapes.

Simple motion activities have characteristic patterns that are visible in the time series representations of the sensor data. These time series representations, such as the 3D accelerations or the Euler angles (roll, pitch and yaw), can be treated as curves and activities can be recognized by matching patterns (shapes) in the curves using curve comparison and alignment techniques.

We transform the sensor signals into cubic B-splines and parametrize the curves with respect to arc length for comparison. We tested our algorithm on the accelerometer data collected at Cleveland State University []. The 3D acceleration signals were segmented at high-level and subject-dependent ‘representative’ curves for the activities were constructed with which test curves were compared and labeled with an overall accuracy rate of 88.46% by our algorithm.

Poorna Talkad Sukumar, K. Gopinath

Sentiment Analysis of Chinese Microblogs Based on Layered Features

Microblogging currently becomes a popular communication way and detecting sentiments of microblogs has received more and more attention in recent years. In this paper, we propose a new approach to detect the sentiments of Chinese microblogs using layered features. Three layered structures in representing synonyms and highly-related words are employed as extracted features of microblogs. In the first layer, “extremely close” synonyms and highly-related words are aggregated into one set while in the second and the third layer, “very close” and “close” synonyms and highly-related words are aggregated respectively. Then in every layer, we construct a binary vector as a feature. Every dimension of a feature indicates whether there are some words in the microblog falling into that aggregated set. These three features provide perspectives from micro to macro. Three classifiers are respectively built from these three features for final prediction. Experiments demonstrate the effectiveness of our approach.

Dongfang Wang, Fang Li

Feature Group Weighting and Topological Biclustering

This paper proposes a new method to weight feature groups for biclustering. In this method, the observations and features are divided into biclusters, based on their characteristics. The weights are introduced to the biclustering process to simultaneously identify the relevance of feature groups in each bicluster. A new biclustering algorithm wBiTM (Weighted Biclustering Topological Map) is proposed. The new method is an extension to self-organizing map algorithm by adding the weight parameter and a new prototype for bicluster. Experimental results on synthetic data show the properties of the weights in wBiTM.

Tugdual Sarazin, Mustapha Lebbah, Hanane Azzag, Amine Chaibi

A Label Completion Approach to Crowd Approximation

Majority vote is one of the most common methods for crowdsourced label aggregation to get higher-quality labels. In this paper, we extend the work of Donmez et al. that estimates majority labels with a small subset of crowdsourcing workers in order to reduce financial and time costs. Our proposed method estimates the majority labels more accurately by completing missing labels to approximate the whole crowds even if some workers do not answer labels. Experimental results show that the proposed method approximates crowds more accurately than the method without label completion.

Toshihiro Watanabe, Hisashi Kashima

Multi-label Linear Discriminant Analysis with Locality Consistency

Multi-label classification is common in many domains such as text categorization, automatic multimedia annotation and bioinformatics, etc. Multi-label linear discriminant analysis (MLDA) is an available algorithm for solving multi-label problems, which captures the global structure by employing the forceful classification ability of the classical linear discriminant analysis. However, some latest studies prove that local geometric structure is crucial for classification. In this paper, we present a new method called

Multi-label Linear Discriminant Analysis with Locality Consistency

(MLDA-LC) which incorporates local structure into the framework of MLDA. Specifically, we employ a graph regularized term to preserve the local structure for multi-label data. In addition, an efficient computing method is also presented to reduce the time and space cost of computation. The experimental results on three benchmark multi-label data sets demonstrate that our algorithm is feasible and effective.

Yuzhang Yuan, Kang Zhao, Hongtao Lu

Hashing for Financial Credit Risk Analysis

Hashing techniques have recently become the trend for accessing complex content over large data sets. With the overwhelming financial data produced today, binary embeddings are efficient tools of indexing big datasets for financial credit risk analysis. The rationale is to find a good hash function such that similar data points in Euclidean space preserve their similarities in the Hamming space for fast data retrieval. In this paper, first we use a semi-supervised hashing method to take into account the pairwise supervised information for constructing the weight adjacency graph matrix needed to learn the binarised Laplacian EigenMap. Second, we train a generalised regression neural network (GRNN) to learn the

k

-bits hash code. Third, the

k

-bits code for the test data is efficiently found in the recall phase. The results of hashing financial data show the applicability and advantages of the approach to credit risk assessment.

Bernardete Ribeiro, Ning Chen

MAP Inference with MRF by Graduated Non-Convexity and Concavity Procedure

In this paper we generalize the recently proposed graduated non-convexity and concavity procedure (GNCCP) to approximately solve the maximum a posteriori (MAP) inference problem with the Markov random field (MRF). Unlike the commonly used graph cuts or loopy brief propagation, the GNCCP based MAP algorithm is widely applicable to any types of graphical models with any types of potentials, and is very easy to use in practice. Our preliminary experimental comparisons witness its state-of-the-art performance.

Zhi-Yong Liu, Hong Qiao, Jian-Hua Su

Two-Phase Approach to Link Prediction

Link prediction deals with predicting edges which are likely to occur in the future. The clustering coefficient of sparse networks is typically small. Link prediction performs poorly on networks having low clustering coefficient and it improves with increase in clustering coefficient. Motivated by this, we propose an approach, wherein, we add relevant non-existent edges to the sparse network to form an auxiliary network. In contrast to the classical link prediction algorithm, we use the auxiliary network for link prediction. This auxiliary network has higher clustering coefficient compared to the original network. We formally justify our approach in terms of Kullback-Leibler (KL) Divergence and Clustering Coefficient of the social network. Experiments on several benchmark datasets show an improvement of upto 15% by our approach compared to the standard approach.

Srinivas Virinchi, Pabitra Mitra

Properties of Direct Multi-Step Ahead Prediction of Chaotic Time Series and Out-of-Bag Estimate for Model Selection

This paper examines properties of direct multi-step ahead (DMS) prediction of chaotic time series and out-of-bag (OOB) estimate of the prediction performance for model selection. Although previous studies of DMS estimation suggest that the DMS technique allows us accuracy improvements from iterated one-step ahead (IOS) prediction. However, it has not considered chaotic time series which has long-term unpredictability as well as short-term predictability, where the boundary of the horizon of long-term and short-term is not known previously. As a result of the model selection, the CAN2 with a large number of units are selected, which is supposed to be useful for avoiding unpredictable data of chaotic time series. We examine the relationship between the OOB prediction and the prediction for the test data, and we suggest that there is a mixed distribution of very small and very big magnitude of prediction errros owing to chaotic time series. We show the effectiveness and the properties of the present method by means of numerical experiments.

Shuichi Kurogi, Ryosuke Shigematsu, Kohei Ono

Multi-document Summarization Based on Sentence Clustering

A main task of multi-document summarization is sentence selection. However, many of the existing approaches only select top ranked sentences without redundancy detection. In addition, some summarization approaches generate summaries with low redundancy but they are supervised. To address these issues, we propose a novel method named Redundancy Detection-based Multi-document Summarizer (RDMS). The proposed method first generates an informative sentence set, then applies sentence clustering to detect redundancy. After sentence clustering, we conduct cluster ranking, candidate selection, and representative selection to eliminate redundancy. RDMS is an unsupervised multi-document summarization system and the experimental results on DUC 2004 and DUC 2005 datasets indicate that the performance of RDMS is better than unsupervised systems and supervised systems in terms of ROUGE-1, ROUGE-L and ROUGE-SU.

Hai-Tao Zheng, Shu-Qin Gong, Hao Chen, Yong Jiang, Shu-Tao Xia

An Ontology-Based Approach to Query Suggestion Diversification

Query suggestion is proposed to generate alternative queries and help users explore and express their information needs. Most existing query suggestion methods generate query suggestions based on document information or search logs without considering the semantic relationships between the original query and the suggestions. In addition, existing query suggestion diversifying methods generally use greedy algorithm, which has high complexity. To address these issues, we propose a novel query suggestion method to generate semantically relevant queries and diversify query suggestion results based on the WordNet ontology. First, we generate the query suggestion candidates based on Markov random walk. Second, we diversify the candidates according the different senses of original query in the WordNet. We evaluate our method on a large-scale search log dataset of a commercial search engine. The outstanding feature of our method is that our query suggestion results are semantically relevant belonging to different topics. The experimental results show that our method outperforms the two well-known query suggestion methods in terms of precision and diversity with lower time consumption.

Hai-Tao Zheng, Jie Zhao, Yi-Chi Zhang, Yong Jiang, Shu-Tao Xia

Sensor Drift Compensation Using Fuzzy Interference System and Sparse-Grid Quadrature Filter in Blood Glucose Control

Diabetes mellitus is a serious chronic condition of the human metabolism. The development of an automated treatment has reached clinical phase in the last few years. The goal is to keep the blood glucose concentration within a certain region with minimal interaction required by the patient or medical personnel. However, there are still several practical problems to solve. One of these would be that the available sensors have significant noise and drift. The latter is rather difficult to manage, because the deviating signal can cause the controller to drive the glucose concentration out of the safe region even in the case of frequent calibration. In this study a linear-quadratic-Gaussian (LQG) controller is employed on a widely used diabetes model and enhanced with an advanced Sparse-grid quadratic filter and a fuzzy interference system-based calibration supervisor.

Péter Szalay, László Szilágyi, Zoltán Benyó, Levente Kovács

Webpage Segmentation Using Ontology and Word Matching

Webpage segmentation is a non trivial task and is a promising research area in the field of computing study. Webpage segments demarcate informative and non-informative content in a webpage through the extraction of text and image. Not only that, segments can also distinguish different types of information between segments. Webpage segmentation is certainly useful in web ranking, classification, and other web mining applications. Segments identification is also useful in page display for constraint limited screen devices such as smart phones, PDAs etc. Recent research focused on using ontology tool to segment a webpage. However, this tool only supports English language. In this paper, we propose a multilingual ontology tool for segmenting webpage. Our tool has shown higher accuracy than existing methods for webpage segmentation.

Huey Jing Toh, Jer Lang Hong

Continuity of Discrete-Time Fuzzy Systems

The purpose of this study is to prove the existence of IF-THEN fuzzy rules which minimize the performance functional of the nonlinear discrete-time feedback control. In our previous study, the problem of fuzzy optimal control was considered as the problem of finding the minimum (maximum) value of the performance function with fuzzy approximate reasoning. This study analyzes a discrete-time system to make numerical simulation of a real model more simple and fast. A continuity of fuzzy approximate reasoning on the compact set of membership functions selected from continuous function space guarantees an optimal control.

Takashi Mitsuishi, Takanori Terashima, Koji Saigusa, Nami Shimada, Toshimichi Homma, Kiyoshi Sawada, Yasunari Shidama

Sib-Based Survival Selection Technique for Protein Structure Prediction in 3D-FCC Lattice Model

Protein Structure Prediction (PSP) is a challenging optimization problem in computational biology. A large number of non-deterministic approaches such as Evolutionary Algorithms (EAs) have been have been effectively applied to a variety of fields though, in the rugged landscape of multimodal problem like PSP, it can perform unsatisfactorily, due to premature convergence. In EAs, selection plays a significant role to avoid getting trapped in local optima and also to guide the evolution towards an optimal solution. In this paper, we propose a new Sib-based survival selection strategy suitable for application in a genetic algorithm (GA) to deal with multimodal problems. The proposed strategy, inspired by the concept of crowding method, controls the flow of genetic material by pairing off the fittest offspring amongst all the sibs (offspring inheriting most of the genetic material from an ancestor) with its ancestor for survival. Furthermore, by selecting the survivors in a hybridized manner of deterministic and probabilistic selection, the method allows the exploitation of less fit solutions along with the fitter ones and thus facilitates escaping from local optima (minima in case of PSP). Experiments conducted on a set of widely used benchmark sequences for 3D-FCC HP lattice model, demonstrate the potential of the proposed method, both in terms of diversity and optimal energy in regard to various state-of-the-art selection methods.

Rumana Nazmul, Madhu Chetty

Tensor Completion Based on Structural Information

In tensor completion, one of the challenges is the calculation of the tensor rank. Recently, a tensor nuclear norm, which is a weighted sum of matrix nuclear norms of all unfoldings, has been proposed to solve this difficulty. However, in the matrix nuclear norm based approach, all the singular values are minimized simultaneously. Hence the rank may not be well approximated. This paper presents a tensor completion algorithm based on the concept of matrix truncated nuclear norm, which is superior to the traditional matrix nuclear norm. Since most existing tensor completion algorithms do not consider of the tensor, we add an additional term in the objective function so that we can utilize the spatial regular feature in the tensor data. Simulation results show that our proposed algorithm outperforms some the state-of-the-art tensor/matrix completion algorithms.

Zi-Fa Han, Ruibin Feng, Long-Ting Huang, Yi Xiao, Chi-Sing Leung, Hing Cheung So

Document Versioning Using Feature Space Distances

The automated analysis of documents is an important task given the rapid increase in availability of digital texts. In an earlier publication, we had presented a framework where the edit distances between documents was used to reconstruct the version history of a set of documents. However, one problem which we encountered was the high computational costs of calculating these edit distances. In addition, the number of document comparisons which need to be done scales quadratically with the number of documents. In this paper we propose a simple approximation which retains many of the benefits of the method, but which greatly reduces the time required to calculate these edit distances. To test the utility of this method, the accuracy of the results obtained using this approximation is compared to the original results.

Wei Lee Woon, Kuok-Shoong Daniel Wong, Zeyar Aung, Davor Svetinovic

Separation and Classification of Crackles and Bronchial Breath Sounds from Normal Breath Sounds Using Gaussian Mixture Model

A computer aided diagnostic system capable of analyzing respiratory sounds can be very helpful in detection of pneumonia, asthma and tuberculosis as the Respiratory sound signal carries information about the underlying physiology of the lungs and is used to detect presence of adventitious lung sounds which are an indication of disease. Respiratory sound analysis helps in distinguishing normal respiratory sounds from abnormal respiratory sounds and this can be used to accurately diagnose respiratory diseases as is done by a medical specialist via auscultation. This process has subjective nature and that is why simple auscultation cannot be relied upon.In this paper we present a novel method for automated detection of crackles and bronchial breath sounds which when coupled together indicate presence and severity of Pneumonia. The proposed system consists of four modules i.e., pre-processing in which noise is filtered out, followed by feature extraction. The proposed system then performs classification to separate crackles and bronchial breath sounds from normal breath sounds.

Ali Haider, M. Daniyal Ashraf, M. Usama Azhar, Syed Osama Maruf, Mehdi Naqvi, Sajid Gul Khawaja, M. Usman Akram

Combined Features for Face Recognition in Surveillance Conditions

This paper addresses the challenging problem of face recognition in surveillance conditions based on the recently published database called SCface. This database emphasizes the challenges of face recognition in uncontrolled indoor conditions. In this database, 4160 face images were captured using five different commercial cameras of low resolution, at three different distances, both lighting conditions and face pose were uncontrolled. Moreover, some of the images were taken under night vision mode. This paper introduces a novel feature extraction scheme that combines parameters extracted from both spatial and frequency domains. These features will be referred to as Spatial and Frequency Domains Combined Features (SFDCF). The spatial domain features are extracted using Spatial Deferential Operators (SDO), while the frequency domain features are extracted using Discrete Cosine Transform (DCT). Principal Component Analysis (PCA) is used to reduce the dimensionality of the spatial domain features while zonal coding is used for reducing the dimensionality of the frequency domain features. The two feature sets were simply combined by concatenation to form a feature vector representing the face image. In this paper we provide a comparison, in terms of recognition results, between the proposed features and other typical features; namely, eigenfaces, discrete cosine coefficients, wavelet subband energies, and Gray Level Concurrence Matrix (GLCM) coefficients. The comparison shows that the proposed SFDCF feature set yields superior recognition rates, especially for images captured at far distances and images captured in the dark. The recognition rates using SFDCF reach 99.23% for images captured by different cameras at the same distance. While for images captured at different distances, SFDCF reaches a recognition rate of 93.8%.

Khaled Assaleh, Tamer Shanableh, Kamal Abuqaaud

Sparse Coding on Multiple Manifold Data

Sparse coding has been widely used in computer vision. While capturing high-level semantics, the independent coding process neglects connections between data points. Some recent methods use Laplacian matrix to learn sparse representations with locality preserving on the manifold. Considering data points may lie in or close to multiple low dimensional manifolds embedded in the high dimensional descriptor space, we use sparse representations to code the local similarity between data points on each manifold and embed this topology to sparse coding algorithm. By keeping the locality of manifolds we can preserve the similarity and separability at the same time. Experimental results on several benchmark data sets show our algorithm is effective.

Hanchao Zhang, Jinhua Xu

Mutual Information Estimation with Random Forests

We present a new method for estimating mutual information based on the random forests classifiers. This method uses random permutation of one of the two variables to create data where the two variables are independent. We show that mutual information can be estimated by the class probabilities of a probabilistic classifier trained on the independent against the dependent data. This method has the robustness and flexibility that random forests offers as well as the possibility to use mixtures of continuous and discrete data, unlike most other approaches for estimating mutual information. We tested our method on a variety of data and found it to be accurate with medium or large datasets yet inaccurate with smaller datasets. On the positive side, our method is capable to estimate the mutual information between sets of both continuous and discrete variables and appears to be relatively insensitive to the addition of noise variables.

Mike Koeman, Tom Heskes

Out-Of-Vocabulary Words Recognition Based on Conditional Random Field in Electronic Commerce

Most previous researches on out-of-vocabulary words recognition are concentrating on traditional areas, while in electronic commerce area this recognition is rarely involved. In this paper, we focus on the out-of-vocabulary words recognition in the field of electronic commerce. Based on the unique characteristics of electronic commerce data, we introduce the Conditional Random Fields (CRFs) into the out-of-vocabulary words recognition, and use the electronic commerce text corpus for the experimental verification. The experimental results show that CRFs in the recognition of out-of-vocabulary words in the field of electronic commerce is effective.

Yanfeng Yang, Yanqin Yang, Hu Guan, Wenchao Xu

Least Angle Regression in Orthogonal Case

LARS(least angle regression) is one of the sparse modeling methods. This article considered LARS under orthogonal design matrix, which we refer to as LARSO. In this article, we showed that LARSO reduces to a simple non-iterative algorithm that is a greedy procedure with shrinkage estimation. Based on this result, we found that LARSO is exactly equivalent with a soft-thresholding method in which a threshold level at the

k

th step is the (

k

 + 1)th largest value of the absolute values of the least squares estimators. For LARSO,

C

p

type model selection criterion can be derived. It is not only interpreted as a criterion for choosing the number of steps/coefficients in a regression problem but also regarded as a criterion for determining an optimal threshold level in LARSO-oriented soft-thresholding method which may be useful especially in non-parametric regression problems. Furthermore, in the context of orthogonal non-parametric regression, we clarified relationship between LARSO with

C

p

type criterion and several methods such as the universal thresholding and SUREshrink in wavelet denoising.

Katsuyuki Hagiwara

Evaluation Protocol of Early Classifiers over Multiple Data Sets

Early classification approaches deal with the problem of reliably labeling incomplete time series as soon as possible given a level of confidence. While developing new approaches for this problem has been getting increasing attention recently, their evaluation are still not thoroughly considered. In this article, we propose a new evaluation protocol for early classifiers. This protocol is generic and does not depend on the criteria used to evaluate the classifiers. Our protocol is successfully applied to 23 publicly available data sets.

Asma Dachraoui, Alexis Bondu, Antoine Cornuéjols

Exploiting Level-Wise Category Links for Semantic Relatedness Computing

Explicit Semantic Analysis(ESA) is an effective method that adopts Wikipedia articles to represent text and compute semantic relatedness(SR). Most related studies do not take advantage of the semantics carried by Wikipedia categories. We develop a SR computing framework exploiting Wikipedia category structure to generate abstract features for texts and considering the lexical overlap between a pair of text. Experiments on three datasets show that our framework could gain better performance against ESA and most other methods. It indicates that Wikipedia category graph is a promising resource to aid natural language text analysis.

Hai-Tao Zheng, Wenzhen Wu, Yong Jiang, Shu-Tao Xia

Characteristic Prediction of a Varistor in Over-Voltage Protection Application

This paper presents a characteristic prediction of a varistor for modeling a varistor in over-voltage protection applications. Variable resistor (Varistor) is one of common devices to protect following electric devices from over-voltage. However, the principle of varistor is still unclear due to its non-linear characteristics between amount of voltage and current. To model the non-linearity, the prediction using adaptive network-based-fuzzy inference system (ANFIS) will be used with several datasets obtained by a high-voltage experiment concerning to the varistor. The result can be used to model the varistor as a voltage clipping device.

Kohei Nagatomo, Muhammad Aziz Muslim, Hiroki Tamura, Koichi Tanno, Wijono

Optimizing Complex Building Renovation Process with Fuzzy Signature State Machines

In contrary to recently built office and commercial buildings, the service life of the traditional European residential houses was not calculated. Some estimations exist about the life span of different types of building constructions, however, these estimations may not reassure the owners of urban-type residential houses that were built before the second world war. A thorough and professional renovation may extend the service life of buildings by decades, the question is how to prepare the most effective renovation procedure.

As a combination of the fuzzy signature structure and the principles of finite-state machine a new formal method is proposed for generating a tool for supporting the renovation planning, concerning the costs and importance of repair. With the support of information obtained from a given pre-war urban-type residential house, the available technical guides and the contractors’ billing database an optimized renovation process of the roof structure is presented as a case study.

Gergely I. Molnárka, László T. Kóczy

News Title Classification with Support from Auxiliary Long Texts

The performance of short text classification is limited due to its intrinsic shortness of sentences which causes the sparseness of vector space model. Traditional classifiers like SVM are extremely sensitive to the features space, thereby making classification performance unsatisfying in short text related applications. It is believed that using external information to help better represent input data would possibly yield satisfying results. In this paper, we target on the problem of news title classification which is an essential and typical member in short text family and propose an approach which employs external information from long text to address the problem the sparseness. Afterwards Restricted Boltzman Machine are utilised to select features and then finally perform classification using Support Vector Machine. The experimental study on Reuters-21578 and Sogou Chinese news corpus has demonstrates the effectiveness of the proposed method.

Yuanxin Ouyang, Yao Huangfu, Hao Sheng, Zhang Xiong

Modelling Mediator Intervention in Joint Decision Making Processes Involving Mutual Empathic Understanding

In this paper an agent model for mediation in joint decision-making processes is presented addressing a disputant-oriented intervention, specifically an education technique. By wielding an education intervention, a mediator can induce a learning process in a disputant. Through this learning process, the disputant may change orientation towards a specific action option. In this way the mediator agent assists two individual social agents in establishing and expressing empathic understanding, as a means to develop solidly grounded joint decisions.

Rob Duell

Backmatter

Weitere Informationen

Premium Partner

    Bildnachweise