Active Media Technology

Understanding of interactions is the critical issue of complex systems. Interactions in physical world are represented by information granules. We propose to model complex systems by interactive intelligent systems (IIS) created by societies of agents. Computations in IIS are based on complex granules (c-granules, for short). Adaptive judgement allows us to reason about c-granules and interactive computations performed on them. In adaptive judgement, different kinds of reasoning are involved such as deduction, induction, abduction or reasoning by analogy as well as intuitive judgement. In modeling of mental parts of c-granules, called information granules (infogranules, for short), we use the approach based on the rough set methods in combination with other soft computing approaches. Issues related to interactions among objects in the physical and mental worlds as well as adaptive judgement belong to the fundamental issues in Wisdom Technology (WisTech). In the paper we concentrate on some basic issues related to interactive computations over c-granules. WisTech was developed over years of work on different real-life projects. It can also be treated as a basis in searching for solutions of problems in such areas as Active Media Technology and Wisdom Web of Things.

This paper proposes a vision-based methodology that recognizes the users’ fingertips so that the users can perform various mouse operations by gestures as well as implements multi-mouse operations. By using the Ramer-Douglas-Peucker algorithm, the system retrieves the coordinates of the finger from the palm of the hand. The system also recognizes the users’ intended operation on the mouse through the movements of recognized fingers. When the system recognizes several palms of hands, it changes its mode to the multi-mouse mode so that several users can coordinate their works on the same screen. The number of mice is the number of recognized palms. In order to implement our proposal, we have employed the Kinect motion capture camera and have used the tracking function of the Kinect to recognize the fingers of users. Operations on the mouse pointers are reflected in the coordinates of the detected fingers. In order to demonstrate the effectiveness of our proposal, we have conducted several user experiments. We have observed that the Kinect is suitable equipment to implement the multi-mouse operations. The users who participated in the experiments quickly learned the multi-mouse environment and performed naturally in front of the Kinect motion capture camera.

Local traditional arts people and the city’s traditional music preservation society have eagerly wished a technology to accurately score traditional music, especially Tsugaru Shamisen. This music will be preserved as scores, which avoids relying solely on the oral transmission of this music to the young performers. At this time, we have experimentally produced an “Electronic Shamisen” equipped with a pick-up microphone attached to each string and automatic scoring equipment, which automatically records scores from the sound source with an agent method.

This paper proposes a system for visualizing individual and collective movement within dense geographical contexts, such as cities and urban neighborhoods. Specifically, we describe a method for creating “spatiotemporal maps” deformed according to personal movement and stasis. We implement and apply a prototype of our system to demonstrate its effectiveness in revealing patterns of spatiotemporal behavior, and in composing maps that more closely correspond to the node-oriented “mental maps” traditionally used by individuals in the act of navigation.

In this paper, we present an investigation of indoor objects positioning using the received Wi-Fi signal strength in the realistic environment with the presence of obstacles. Wi-Fi RSS based positioning is a promising alternative to other techniques for locating indoor objects. Two factors may lead to the low Wi-Fi RSS positioning accuracy: the existence of moving obstacles, and the limited number of available anchor nodes. We propose a novel approach to locating a target object in a given area by introducing a hidden factor for a reduced form of probabilistic estimator. This estimator is unbiased with the scalability in field size. With the selection of a Gaussian prior on this hidden factor characterizing the effects of RSS drop introduced by obstacles, we convert the positioning prediction into a maximum a posteriori problem, then apply expectation-maximization algorithm and conjugate gradient optimization to find the solution. Simulations in various settings show that the proposed approach presents better performance compared to other state-of-the-art RSS range-based positioning algorithms.

Studies of human-computer interaction have been broadened and deepened and powerful, novel gestural interfaces are of special interest in this field. This paper concerns the motion of feet on a floor sensor device. We have been investigating footstep tracking of individuals walking on the device and have presented a particle filter-based framework. In this paper, we present a trial of extended system facilities, allowing the system to identify pairs of footsteps of individuals walking on the floor sensor device. Three gait characteristics, stride length, footstep direction, and relative pressure values in foot regions, are considered. In addition, an implementation of the improved particle filter framework with Walker’s alias method is also described for speeding execution.

In this paper, we propose a method to detect failure of learning of students using quiz responses in a course management system Moodle. Failure of learning is defined as a situation in which the correct answer rate of a learning topic is significantly lower than the correct answer rate of other topics. In the research, the researchers identified the presence or absence of failure of learning in actual classes to evaluate the usefulness of the proposed method. The results revealed that students checked the quiz results significantly more in the experimental group than in the control group, and that more instruction was given to the experimental group.

The purpose of this paper is two-fold. First, it seeks to introduce the conceptual prototype of SpamCombat, a Web application that helps combat spam through gamification of community policing. Second, it attempts to evaluate SpamCombat by identifying factors that can potentially drive users’ behavioral intention to adopt. A questionnaire seeking quantitative and qualitative responses was administered to 120 participants. The results indicate that behavioral intention to adopt SpamCombat is generally promising. Most participants appreciated the novelty of SpamCombat in supporting community policing to promote a spam-free cyber space. However, participants felt that using SpamCombat could be time-consuming.

Learning from demonstrations (LfD) is receiving more attention recently as an important modality for teaching robots and other agents new skills by untrained users. A successful LfD system must tackle several problems including the decision about what and whom to imitate but, ultimately, it needs to reproduce the skill it learned solving the

how to imitate

problem. One promising approach to solving this problem is using Gaussian Mixture Modeling and Gaussian Mixture Regression for reproduction. Most available systems that utilize this approach rely on kinesthetic teaching or require the attachment of special markers to measure joint angles of the demonstrator. This bypasses the correspondence problem which is accounting for the difference in the kinematic model of the demonstrator and the learner. This paper presents a closed-form analytic solution to the correspondence problem for an upper-body of a humanoid robot that is general enough to be applicable to many available humanoid robots and reports the application of the method to a pose copying task executed by a NAO robot using Kinect recorded data of human demonstrations.

Learning and utilizing a pool of features for a given data is important to achieve better performance in data analysis. Since many real world data can be represented as a non-negative data matrix, Non-negative Matrix Factorization (NMF) has recently become popular to deal with data under the non-negativity constraint. However, when the number of features is increased, the constraint imposed on the features can hinder the effective utilization of the learned representation. We conduct extensive experiments to investigate the effectiveness of several state-of-the-art NMF algorithms for learning and utilizing a pool of features over document datasets. Experimental results revealed that coping with the non-orthogonality of features is crucial to achieve a stable performance for exploiting a large number of features in NMF.

Various research efforts have been invested in machine learning and data mining for finding out patterns from data. However, even when some knowledge may be learned in one domain, it is often difficult to re-use it for another domain with different characteristics. Toward effective knowledge transfer between domains, we proposed a transfer learning method based on our transfer hypothesis that two domains have similar feature spaces. A graph structure called a topic graph is constructed by using the learned features in one domain, and the graph is used as a regularization term. In this paper we present a theoretical analysis of our approach and prove the convergence of the learning algorithm. Furthermore, the performance evaluation of the method is reported over document clustering problems. Extensive experiments are conducted to compare with other transfer learning algorithms. The results are encouraging, and show that our method can improve the performance by transferring the learned knowledge effectively.

VSF–Network, Vibration Synchronizing Function Network, is a hybrid neural network combining Chaos Neural Network and hierarchical neural network. VSF–Network is designed for symbol learning. VSF–Network finds unknown parts of input data by comparing to stored pattern and it learns unknown patterns using unused part of the network. New patterns are learned incrementally and they are stored as sub-networks . Combinations of patterns are represented as combinations of the sub-networks. In this paper, the two theoretical backgrounds of VSF–Network are introduced. At the first, an incremental learning framework with Chaos Neural Networks is introduced. Next, the pattern recognition with the combined with symbols is introduced. From the viewpoints of9 differential topology and mixture distribution, the combined pattern recognition by VSF-Network is explained. Through an experiment, both the incremental learning capability and the pattern recognition with pattern combination are shown.

This paper presents an approach toward robust and fast Two-Dimensional Linear Discriminant Analysis (2DLDA). 2DLDA is an extension of Linear Discriminant Analysis (LDA) for 2-dimensional objects such as images. Linear transformation matrices are iteratively calculated based on the eigenvectors of asymmetric matrices in 2DLDA. However, repeated calculation of eigenvectors of asymmetric matrices may lead to unstable performance. We propose to use simultaneous diagonalization of scatter matrices so that eigenvectors can be stably calculated. Furthermore, for fast calculation, we propose to use approximate decomposition of a scatter matrix based on its several leading eigenvectors. Preliminary experiments are conducted to investigate the effectiveness of our approach. Results are encouraging, and indicate that our approach can achieve comparative performance with the original 2DLDA with reduced computation time.

Under the condition of multi-databases, a novel algorithm of facial expression recognition was proposed to improve the robustness of traditional semi-supervised methods dealing with individual differences in facial expression recognition. First, the regions of interest of facial expression images were determined by face detection and facial expression features were extracted using Linear Discriminant Analysis. Then Transfer Learning Adaptive Boosting (TrAdaBoost) algorithm was improved as semi-supervised learning method for multi-classification. The results show that the proposed method has stronger robustness than the traditional methods, and improves the facial expression recognition rate from multiple databases.

Nowadays, biclustering problem is still an intractable problem. But in time series expression data, the clusters can be limited those with contiguous columns. This restriction makes biclustering problem to be tractable problem. However existing contiguous column biclustering algorithm can only find the biclusters which have the same value for each column in biclusters without error tolerance. This characteristic leads the algorithm to overlook some patterns in its clustering process. We propose a suffix tree based algorithm that allows biclusters to have inconsistencies in at most k contiguous column. This can reveals previously undiscoverable biclusters. Our algorithm still has tractable run time with this additional feature.

Teams are ranked to show their authority over each other. The International Cricket Council (ICC) ranks the cricket teams using an ad-hoc points system entirely based on the winning and losing of matches. In this paper, adoptions of PageRank and h-index are proposed for ranking teams to overcome the weakness of ICC ad-hoc point system. The intuition is to get more points for a team winning from a stronger team than winning from a weaker team by considering the number of runs and wickets also in addition to just winning and losing matches. The results show that proposed ranking methods provide quite promising insights of one day and test team rankings.

Feature selection is a well known and studied technique that aims to solve “the curse of dimensionality” and improve performance by removing irrelevant and redundant features. This paper highlights some well known approaches to filter feature selection, information theory and rough set theory, and compares a recent fitness function with some traditional methods. The contributions of this paper are two-fold. First, new results confirm previous research and show that the recent fitness function can also perform favorably when compared to rough set theory. Secondly, the measure of redundancy that is used in traditional information theory is shown to damage the performance when a similar approach is applied to the recent fitness function.

In this paper, a method for constructing transferable “web” and “clickstream” prediction models based on sequential pattern evaluation indices is proposed. To predict end points, click streams are assumed as sequential data. Further, a sequential pattern generation method is applied to extract features of each click stream data. Based on these features, a classification learning algorithm is applied to construct click stream end point prediction models. In this study, the evaluation indices for sequential patterns are introduced to abstract each clickstream data for transferring the constructed predictive models between different periods. This method is applied to a benchmark clickstream dataset to predict the end points. The results show that the method can obtain more accurate predictive models with a decision tree learner and a classification rule learner. Subsequently, the evaluation of the availability for transferring the predictive morels between different periods is discussed.

Recommender systems in online marketing websites like Amazon.com and CDNow.com suggest relevant services and favorite products to customers. In this paper, we proposed a novel hypergraph-based kernel computation combined with

k

nearest neighbor (

k

NN) to predict ratings of users. In this method, we change regular definition style of hypergraph diffusion kernel. Our comparative studies show that our method performs better than typical

k

NN, which is simple and appropriate for online recommending applications.

Social tagging systems extend recommender systems from the pair (user, item) to (user, item, tag). This paper discusses the framework of similarity measure on (user, item, tag) from qualitative and quantitative perspectives. The qualitative measure makes use of the preference structure relation on (user, item, tag), and the quantitative measure makes use of reflection on (user, item, tag). The

k

nearest neighbors and reverse

k

′ nearest neighbors are used to generate recommendations.

Urban computing aims to enhance both human life and urban environment smartly by deeply understanding human behavior occurring in urban area. Nowadays, mobile phones are often used as an attractive option for large-scale sensing of human behavior, providing a source of real and reliable data for urban computing. But analyzing the data also faces some challenges (e.g., the related data is heterogeneous and very big), and the general approaches cannot deal with them efficiently. In this paper, aiming to tackle these challenges and conduct urban computing efficiently, we propose a data integration model for the multi-source heterogeneous data related to mobile phones by using semantic technology and develop a semantic mobile data management system.

This paper treats methodology to build linked data from the relationships between facial action units and their states as emotional parameters for the facial emotion recognition. In this paper, the authors are especially focusing on building action unit-based linked data because it will be possible not only to use the data for the facial emotion recognition but also to enhance the usefulness of the data by merging them with other linked data. Although in general, the representation as linked data seems to make the accuracy of the facial emotion recognition lower than others, in practically the proposed method that uses action unit-based linked data has almost the same accuracy for the facial emotion recognition as those of other approaches like using Artificial Neural Network and using Support Vector Machine.

Visual information, especially in the form of images, is becoming increasingly important, and consequently there is a rising demand for effective tools to perform online image search. However, image search engines such as Google Images, are based on the text surrounding the images rather than the images themselves. At the same time, while the employed keyword-based search provides a basic level of filtering, it is not sufficient to handle large search results. Image database visualisation, which provides a visual overview of an image collection, could be applied to the retrieved images, but the associated overheads, both in terms of bandwidth and computational complexity, are prohibitive.

In this paper, we introduce an image browsing system that does not suffer from these drawbacks. In particular, we construct an interactive image database navigation application that uses the Huffman tables available in the JPEG headers of Google Images thumbnails directly as image features, and projects images onto a 2-dimensional visualisation space based on principal component analysis derived from the Huffman entries. Images are dynamically placed into a grid structure and organised in a tree-like hierarchy for visual browsing. Since we utilise information only from the JPEG header, the requirement in terms of bandwidth is very low, while no explicit feature calculation needs to be performed, thus allowing for interactive browsing of online image search results.

This paper proposes a dominant-submissive agent model on bounded confidence opinion dynamics under an emergency environment. In the proposed model, environmental noises and opinion leaders are involved in the collective opinion formation. A series of computer simulations demonstrate that environmental noises have a great impact on the collective opinion evolution. The interactions among individuals are strengthened as the variances of the environmental noises increase, and then a global group behavior emerge with a higher probability. On the other hand, the influence of opinion leaders on the collective opinion dynamics is limited. Firstly, when the fraction of opinion leaders is fixed in the social network, the number of agents following the opinion leaders decreases as the variance of the environmental noise exceeds a certain threshold. Secondly, the number of agents following the opinion leaders does not change obviously as the fraction of opinion leaders increases under a constant noisy environment.

Subjectivity analysis investigates attitudes, feelings, and expressed opinions about products, services, topics, or issues. As the basic task, it classifies a text as subjective or objective. While subjective text expresses opinions about an object or issue using sentiment expressions, objective text describes an object or issue considering their facts. The presence of sentiment terms such as adjectives, nouns and adverbs in products reviews usually implicates their subjectivity, but for comments about social issues, it is more complicated and sentiment phrases and patterns are more common and descriptive. This paper proposes a lexical-syntactical structure for subjective patterns for subjectivity analysis in social domains. It is employed and evaluated for subjectivity and sentiment classification at the sentence level. The proposed method outperforms some similar works. Moreover, its reasonable F-measure implicates its usability in applications like sentiment summarization and opinion question answering.

We examined the possible implication of two different technological tools, the touch screen and the keyboard, on cross-modal interaction in writing. To do this, we revisit experiments (e.g. [1]) that investigated the recovery of spatial iconicity in semantic judgment and applied them in writing to dictation. In the present experiment participants had to type or to handwrite on a touchscreen, in the upper part or in the lower part of the screen, words whose referents are typically associated with the top or the bottom part of space. In this way congruent (e.g. cloud at the top of the screen) or incongruent (e.g. cloud at the bottom of the screen) conditions were created. The hypothesis was that incongruent conditions give rise to a delay in starting to write more pronounced for touch screen session than for the keyboard one. Results are discussed in terms of embodied cognition theory.

This study aimed to develop a novel treatment method for tinnitus using phase-shift sound stimulation. We performed physical audio signal processing to create simulated sound stimuli resembling subjective tinnitus. The preliminary study utilized two tinnitus models representing different origins of tinnitus, and in each model the simulated tinnitus sound was presented simultaneously with a phase-shifted sound. We then measured audio brainstem response wave latencies; wave latency prolongation served as an evaluation index. The main study involved subjects with tinnitus but no underlying disease. To modulate the perception of tinnitus, an oscillator was used to identify tinnitus frequency and produce sound output that was then phase shifted. Preliminary study results indicated that excitation of the nerve impulse by an additional sound can modulate coding of preceding tinnitus information in the auditory brainstem. The experimental study demonstrated the reproducibility of time delays. Both results suggest the clinical usefulness of this treatment method.

The

virtual forms

present dynamically generated visual images containing information that influences a users behavior and thinking. In a typical way, adding a display to show visual expressions or projecting some information on an artifact offers computational visual forms on the existing daily artifacts. Using

virtual forms

is a very promising way to enhance artifacts surrounding us, and to make our daily life and business richer and more enjoyable. We believe that incorporating fictional stories into

virtual forms

offers a new possibility for enriching user experiences. In particular, integrating fictional stories into our daily activities through transmedia storytelling is a promising approach. Transmedia storytelling enables

virtual forms

to be employed everywhere to immersively integrate fictional stories into our daily activities. If we can design attractive

virtual forms

in a structured way, it will become easy to enrich user experiences. Currently, the design framework for

virtual forms

is not well defined. The framework needs to take into account the semiotic aspect of a

virtual form

. One key factor, in particular, is how strongly we believe in the reality of a fictional story within the

virtual form

. In this paper, we show the extracted insights discussed in the workshops and present some design implications for designing

virtual forms

that integrate fictional stories into our daily activities.

E-mail-based communication and collaboration are important to organizational work. In order to help the multi-tasking knowledge worker, a task management-based software environment requires a support mechanism of an automated user operation in addition to a support function that manages task resources. In this paper, we propose a task context model that manages task-related e-mail messages and their resources for the purpose of reusing them. In addition, we describe a task context model-based user assist functions that allows users to send or reply to e-mail quickly and efficiently and that automatically extracts data from e-mail messages. To validate the task context model, we implement the prototype system and describe its experimental results.

Extraction and management of technical terminologies become an important process in the business intelligence. To do this, historic methods have a focus on calculating weight values and selecting top

n

terminologies according to the values for the cores that represent given scientific documents. These terminologies selected through those methods can be used as important clues for business intelligence services such as technology trend analysis, potential market discover, and so on however the terminologies extracted from the documents do not mean the technologies of the organizations publishing the documents. Therefore, our research is based on a fundamental that there are only a few technologies an organization participates in directly even though a scientific document of the organization contains various technical terminologies. In this paper, to enhance the quality of business intelligence services, we propose a method to select core technologies of an organization and utilize semantic networks of technical terminologies of a given scientific document and we suggest its possibility through simple experimental evaluation.

Numerous linguistic resources are readily available in area of expertise due to the development of wireless devices such as smart-phones and the internet. To select useful information from the massive amount of the data, many systems using semantic web technologies have been developed. In order to build those systems, data collection and natural language processing are essential. However, most of those systems do not consider software integration and the data required by the processes used. In this paper, we propose a system, entitled uLAMP which integrates software and data related to natural language processing. In terms of economics, the cost is reduced by preventing duplicated implementation and data collection. On the other hand, data and software usability are increasing in terms of management requirements. In addition, for the evaluation of the uLAMP usability and effectiveness of uLAMP, a user survey was conducted. Through this evaluation, the advantages of the currentness of data and the ease of use were found.

Existing visualizing methods for big ontology data have many problems. To solve the problems and visualize big ontology data efficiently, we used Hadoop framework, which is for distributed processing across clusters for handling large dataset. The system that we devised is made up of three parts-a data server, a visualization server, and user devices. First of all, The data server preprocesses big data, and the visualization server processes the outputs for visualizing them and transform the outputs to match web standard. The data server and the visualization server use Hadoop framework. User devices have web browsers. Through web browsers, users can be provided with the visualization results by the visualization server We processed DBpedia ontology and visualized the data. In this paper, we will introduce a method for processing and visualizing DBPedia ontology. And we will show the performance of the method by measuring execution time and the experimental results of the visualizing process.

Social Learning as a new concept of learning model emphasizes an individual’s activity and formation of relationships with other people. On the contrary, traditional recommendation system provides a target user with the appropriate recommendation information after analyzing a user’s preference based on the user’s profiles and rating histories. These kinds of systems need to modify recommendation algorithm; these traditional recommendation systems are limited to only two attributes - user profiles and rating histories – that includes the problem of recommendation reliability and accuracy. In this paper, we present a user-context based collaborative filtering (UCCF) using user-context and social relationships. The UCCF analyzes user-context and social relationships, and generates a similar user group which uses the user’s recommendation score from similar user groups. The UCCF reflects strong ties of users who have similar tendency and improves reliability and accuracy of the content and expert recommendation system.

This paper proposes an automatic animation composition system based on six databases, which are defined through analyzing a Japanese folktale animation movie. The system can automatically translate a text-based event representation (simple case frame) into an animation script (TVML). We show that our proposed system can compose several TVML scripts, which can represent animations of events as parts of a story.

This research discusses the generation method of the narrative text linked with geographic space data. First, analysis methods for the existing folk tale text are described. It is clarified that there is geographic bias in the narrative text. Next, the technique for using geographic space data for the sightseeing tour is described. The sightseeing tour is an action for the tourist to touch the narrative text in the local area while moving. We maintained the place that related to the text as geographic space data. The text including place information code can be plotted on the digital map. We propose a new expression technique of the story text by integrating these methods.

This paper discusses some methods for generalizing our Propp-based story generation mechanism. The mechanism has the following aspects: as the development of a system based on the literary theory by Propp and as the use in a more comprehensive architecture of narrative generation called the integrated narrative generation system. Considering the latter aspect especially, the generalization beyond the restriction of Propp’s theory will become an important issue for the future development. The first half of this paper will introduce overviews of the Propp-based mechanism and the integrated narrative generation system. Then in the latter half, we will present four methods for the generalization.

In this paper, we formulate an optimal weight design (OWD) problem of a 23 stories frame structure for a constrained relative story displacement as a statically indeterminate structure problem and solve it directly by keeping the constraints based on an improved genetic algorithm (GA). We discuss the efficiency between the proposed method and the discredited optimum criteria methods.

Transportation problem is a typical combinatorial problem. We aim at the search capacity of the solution by using a genetic algorithm with a the Bicriteria fixed Charge Transportation problem, which is an extension of the traditional transportation problem. In this paper, we improve the technique of Teramatu. In particular we propose new crossover operation for the genetic algorithm. Comparison with other methods is performed and the validity of the proposed method is shown.

In data analysis, we must be conscious of the probability density function of population distribution. Then it is a problem why the probability density function is expressed.

The estimation of a probability density function based on a sample of independent identically distributed observations is essential in a wide range of applications. The estimation method of probability density function – (1)a parametric method (2)a nonparametric method and (3)a semi-parametric method etc. – it is. In this paper, GMM problem is taken up as a semi-parametric method and We use a wavelet method as a powerful new technique. Compactly supported wavelets are particularly interesting because of their natural ability to represent data with intrinsically local properties.

This paper describes a job-shop scheduling problem (JSP) of processing product subject to no delay time job. It is one objective model of the minimum delivery delay time.In this paper, the effectiveness we the numerical experiments using a benchmark problem to improve the solution accuracy and decrease execution time by adding a method to generate gene and new approach, we introduce a search method for the algorithm shorter delivery times and further there to verify.

Recently, genetic algorithms (GA) have received considerable attention regarding their potential as a combinatorial optimization for complex problems and have been successfully applied in the area of various engineering. We will survey recent advances in hybrid genetic algorithms (HGA) with local search and tuning parameters and multiobjective HGA (MO-HGA) with fitness assignments. Applications of HGA and MO-HGA will introduced for flexible job-shop scheduling problem (FJSP), reentrant flow-shop scheduling (RFS) model, and reverse logistics design model in the manufacturing and logistics systems.

The large-scale unordered sentences are springing up on the web since the massive novel web social Medias have emerged. Although those unordered sentences have rich information, they only provide users with incoherent information service because they have loose semantic relations. Users usually expect to obtain semantic coherent information service when they are facing massive unordered sentences. Unfortunately, general web search engines are not applicable to such issue, because they only return a flat list of unordered web pages based on keywords. In this paper, we propose a novel semantic coherence based intelligent search system. The search system can provide semantic coherence based search service, which includes choosing semantic coherent sentences and ranking the sentences by a semantic coherent way. When a user enters some semantic incoherent sentences as queries, our system can return a semantic coherent paragraph as search results. The process is demonstrated by a prototypical system and experiments are conducted to validate its correctness. The results of experiments have shown that the system can distinguish semantic coherent sentences from others and rank the sentences by a semantic coherent way with higher accuracy.

Data replication can improve the performance and availability for applications, and when it is employed by big data applications, it has to solve the challenges posed by big data applications, i.e., offering scalability and varying consistency levels. In this paper, we design and implement a data replication framework Pyxis+, whereby replication-aware applications can be developed in a rapid and convenient way. Pyxis+ allows the applications to register different consistency levels and automatically switches the consistency levels according to the change of requirements and performance. Meanwhile, on the basis of the consistency guarantees, Pyxis+ takes advantage of the consistent hashing technology to improve the scalability of data access. Simulation experimental results show that Pyxis+ can obtain relatively stable throughputs and response time by adding or removing replica managers while facing the increase of user requests.

The paper investigates the use of support vector machines (SVM) in classifying Matrix-Assisted Laser Desorption Ionisation (MALDI) Time Of Flight (TOF) mass spectra. MALDI-TOF screening is a simple and useful technique for rapidly identifying microorganisms and classifying them into specific subtypes. MALDI-TOF data presents data analysis challenges due to its complexity and inherent data uncertainties. In addition, there are usually large mass ranges within which to identify the spectra and this may pose problems in classification. To deal with this problem, we use Wavelets to select relevant and localized features. We then search for best optimal parameters to choose an SVM kernel and apply the SVM classifier. We compare classification accuracy and dimensionality reduction between the SVM classifier and the SVM classifier with wavelet-based feature extraction. Results show that wavelet-based feature extraction improved classification accuracy by at least 10%, feature reduction by 76% and runtime by over 80%.

Content-based image retrieval (CBIR) has been widely studied in recent years. CBIR usually employs feature descriptors to describe the concerned characters of images, such as geometric descriptor and texture descriptor. Many texture descriptors in texture analysis and image retrieval are based on the so-called Local Binary Pattern (LBP) technique. However, LBP lacks of the spatial distribution information of texture features. In this paper, we aim at improving the traditional LBP and present a novel texture feature descriptor for CBIR called Multi-Scale Local Spatial Binary Patterns (MLSBP). MLSBP integrates LBP with spatial distribution information of gray-level variation direction and gray-level variation between the referenced pixel and its neighbors. In addition, MLSBP extracts the texture features from images on different scale levels. We conduct experiments to compare the performance of MLSBP with five competitors including LBP, Uniform LBP (ULBP), Completed LBP (CLBP), Local Ternary Patterns (LTP), and Local Tetra Patterns (LTrP). Also three benchmark image databases are used in the measurement, which are the Bradotz Texture Database (DB1), the MIT VisTex Database (DB2), and the Corel 1000 Database (DB3). The experimental results show that MLSBP is superior to the competitive algorithms in terms of precision and recall.