Vol. 31 · No. 2

Machine translation systems can be classified into rule-based and corpus-based approaches, in terms of their core methodology. Since both paradigms have been largely used during the last years, one of the aims in the research community is to know how these systems differ in terms of translation quality. To this end, this paper reports a study and comparison of several specific Catalan-Spanish machine translation systems: two rule-based and two corpus-based (particularly, statistical-based) systems, all of them freely available on the web. The translation quality analysis is performed under two different domains: journalistic and medical. The systems are evaluated by using standard automatic measures, as well as by native human evaluators. In addition to these traditional evaluation procedures, this paper reports a novel linguistic evaluation, which provides information about the errors encountered at the orthographic, morphological, lexical, semantic and syntactic levels. Results show that while rule-based systems provide a better performance at orthographic and morphological levels, statistical systems tend to commit less semantic errors. Furthermore, results show all the evaluations performed are characterised by some degree of correlation, and human evaluators tend to be specially critical with semantic and syntactic errors.

Because virtual computing platforms are dynamically changing, it is difficult to build high-quality intrusion detection system. In this paper, we present an automated approach to intrusions detection in order to maintain sufficient performance and reduce dependence on execution environment. We discuss a hidden Markov model strategy for abnormality detection using frequent system call sequences, letting us identify attacks and intrusions automatically and efficiently. We also propose an automated mining algorithm, named AGAS, to generate frequent system call sequences. In our approach, the detection performance is adaptively tuned according to the execution state every period. To improve performance, the period value is also under self-adjustment.

We present a new dependency parsing algorithm based on the decomposition of large sentences into smaller units such as clauses and intraclausal coordinations. For the identification of these units, new methods combining machine learning techniques and heuristic rules were developed. The algorithm was evaluated on the Slovene dependency treebank text corpus. Compared to the MSTP parser, currently the most accurate for Slovene, parsing accuracy was improved by 1.27 percentage points, which equals 6.4

Association rule mining is an active data mining research area. Recent years have witnessed many efforts on discovering fuzzy associations. The key strength of fuzzy association rule mining is its completeness. This strength, however, comes with a major drawback to handle large datasets. It often produces a huge number of candidate itemsets. The huge number of candidate itemsets makes it ineffective for a data mining system to analyze them. In the end, it produces a huge number of fuzzy associations. This is particularly true for datasets whose attributes are highly correlated. The huge number of fuzzy associations makes it very difficult for a human user to analyze them. Existing research has shown that most of the discovered rules are actually redundant or insignificant. In this paper, we propose a novel technique to overcome these problems; we are preprocessing the data tuples by focusing on similar behaviour attributes and ontology. Finally, the efficiency and advantages of this algorithm have been proved by experimental results.

Consider an orthogonal grid of streets and avenues in a Manhattan-like city populated by stationary sensor modules at some intersections and mobile robots that can serve as relays of information that the modules exchange, where both module-module and module-robot communication is limited to a straight line of sight within the grid. The robots are oblivious and move asynchronously. We present a distributed algorithm that, given the sensor locations as input, moves the robots to suitable locations in the grid so that a connected network of all modules is established. The number of robots that the algorithm uses is worst case optimal.

Swarm Intelligence (SI) is an innovative distributed intelligent paradigm whereby the collective behaviors of unsophisticated individuals interacting locally with their environment cause coherent functional global patterns to emerge. Although the swarm algorithms have exhibited good performance across a wide range of application problems, it is difficult to analyze the convergence. In this paper, we discuss the dynamic trajectory and convergence of the swarm intelligent model, namely the particle swarm algorithm. We explore the tradeoff between exploration and exploitation using differential analysis and Laplace transform. The trajectories are parsed into first-order inertial element and second-order oscillation element. Their transfer functions are derived, and the trajectories are described in explicit time functions. The first-order inertial element is helpful to maintain the trajectory's stability and algorithm convergence, while the second-order oscillation element trends to explore some new search spaces for the better solutions. The convergence regions of the swarm system are analyzed using the spectral radius and Lyapunov second theorem on stability.

Out-of-order tuples in continuous data streams may cause inaccurate query results since conventional window operators generally discard those tuples. Existing approaches use a buffer to fix disorder in stream tuples and estimate its size based on the maximum network delay seen in the streams. However, they do not provide a method to control the amount of tuples that are not saved and discarded from the buffer, although users may want to keep it within a predefined error bound according to application requirements. In this paper, we propose a method to estimate the buffer size while keeping the percentage of tuple drops within a user-specified bound. The proposed method utilizes tuples' interarrival times and their network delays for estimation, whose parameters reflect real-time stream characteristics properly. Based on two parameters, our method controls the amount of tuple drops adaptively in accordance with fluctuated stream characteristics and keeps their percentage within a given bound, which we observed through our experiments.

Testing of a deeply embedded mixed-signal core in a System-on-Chip (SoC) is a challenging issue due to the communication bottleneck in accessing the core from external automatic test equipment. Consequently, in many cases the preferred approach is built-in self-test (BIST), where the major part of test activity is performed within the unit-under-test and only final results are communicated to the external tester. IEEE Standard 1500 provides efficient test infrastructure for testing digital cores; however, its applications in mixed-signal core test remain an open issue. In this paper we address the problem of implementing BIST of a mixed-signal core in a IEEE Std 1500 test wrapper and discuss advantages and drawbacks of different test strategies. While the case study is focused on histogram based test of ADC, test strategies of other types of mixed-signal cores related to trade-off between performance (i.e., test time) and required resources are likely to follow similar conclusions.

Monitoring services are an essential component of large-scale computing infrastructures due to providing information which can be used by humans as well as applications to closely follow the progress of computations, to evaluate the performance of ongoing computing, etc. However, the users are usually left alone with performance measurements as to the interpreting and detecting of execution flaws. In this paper we present an approach to the performance monitoring of distributed applications based on semantic information about the monitored objects involved in the application execution. This allows to automate the guidance on what to measure further to come to a source of performance flaws as well to enable reacting on interesting events, e.g. on exceeding SLA parameters. Our research comprises the implementation of a robust system with semantics, which is not biased to an underlying ``physical'' monitoring system, giving the end user the power of intelligent monitoring functionality as well as the independence of the heterogeneity of distributed infrastructures.

In the Web 2.0 era, people not only read web contents but create, upload, view, share and evaluate all contents on the web. This leads us to introduce a new type of social network based on user activity and content metadata. We notice that we can determine the quality of related contents using this new social network. Based on this observation, we introduce a user evaluation algorithm for user-generated video sharing website. First, we make a social network of users from video contents and related social activities such as subscription, uploading or favorite. We then use a modified PageRank algorithm to compute user reputation from the social network. We re-calculate the content scores using user reputations and compare the results with a standard BM25 result. We apply the proposed approach to YouTube and demonstrate that the user reputation is closely related to the number of subscriptions and the number of uploaded contents. Furthermore, we show that the new ranking results relied on the user reputation is better than the standard BM25 approach by experiments.

e-Science applications increasingly require both computational power and storage resources, currently supported with a certain level of quality. Since in the grid and cloud environments, where we can execute the e-Science applications, heterogeneity of storage systems is higher than that of computational power resources, optimization of data access defines one of challenging tasks nowadays. In this paper we present our approach to management of data access in the grid environment. The main issue is to organize data in such a way that users requirements in the form of QoS/SLA are met. For this purpose we make use of a storage monitoring system and a mass storage system model – CMSSM. The experiments are performed in the PL-Grid environment.