Vol. 34 · No. 6

The growing importance of security operations in urban terrain has triggered many attempts to address the perceived gaps in the readiness of security forces for this type of combat. One way to tackle the problem is to employ simulation techniques. Simulations are widely used to support both mission rehearsal and mission analysis, but these two applications tend to be seen as distinctly separate. We argue that integrating them in a unified framework can bring significant benefits for end-users. We perform a structured walk-through of such a unified system, in which a novel approach to integration through the behaviour cloning enabled the system to capture the operational knowledge of security experts, which is often difficult to express verbally. This capability emerged as essential for the operation of the integrated system. We also illustrate how the interplay between the system components for the mission analysis and mission rehearsal is realized.

We present an agent-based model of a simple endogenous-money economy. The model simulates agents representing individual persons who can work, consume, invent new products and related production technologies, apply for a loan from the bank and start up a business. Through the interaction of persons with the firms, we simulate the production of goods, consumption and labour market. In order to achieve a significant level of realism of the simulations, the firms are modelled as adaptive agents using an effective reinforcement learning approach in continuous space. This setting allows us to explore how an endogenous-money economy can be built up from scratch, as an emergent property of actions and interactions among heterogeneous agents once money is injected into a non-monetary self-production (or barter) economy. In the paper, we first empirically investigate the learning capability of the firm agents. Then, we discuss the results of some computational experiments under different significant scenarios.

This paper explores the application of structured learning methods (SLMs) to word sense disambiguation (WSD). On one hand, the semantic dependencies between polysemous words in the sentence can be encoded in SLMs. On the other hand, SLMs obtained significant achievements in natural language processing, and so it is a natural idea to apply them to WSD. However, there are many theoretical and practical problems when SLMs are applied to WSD, due to characteristics of WSD. Beginning with the method based on hidden Markov model, this paper proposes for the first time a comprehensive and unified solution for WSD based on maximum entropy Markov model, conditional random field and tree-structured conditional random field, and reduces the time complexity and running time of the proposed methods to a reasonable level by beam search, approximate training, and parallel training. The update of models brings performance improvement, the introduction of one step dependency improves performance by 1–5 percent, the adoption of non-independent features improves performance by 2–3 percent, and the extension of underlying structure to dependency parsing tree improves performance by about 1 percent. On the English all-words WSD dataset of Senseval-2004, the method based on tree-structured conditional random field outperforms the best attendee system significantly. Nevertheless, almost all machine learning methods suffer from data sparseness due to the scarcity of sense tagged data, and so do SLMs. Besides improving structured learning methods according to the characteristics of WSD, another approach to improve disambiguation performance is to mine disambiguation knowledge from all kinds of sources, such as Wikipedia, parallel corpus, and to alleviate knowledge acquisition bottleneck of WSD.

An energy efficient way of disseminating XML data to several mobile clients is broadcast. Information such as alert on emergencies, election results and sporting event results can be of interest to large number of mobile clients. Since eXtensible Markup Language (XML) is widely used for information exchange, wireless information services require an energy efficient XML data dissemination. XML Path (XPath) represents selective data required by mobile clients. XPath query processing involves two performance metrics, namely tune-in time and access time. In this paper, we propose a novel structure for streaming XML data called Path Stream Group Level (PSGL) node by exploiting the tree structure of XML document. It possesses various small indices such as level, child, sibling, attribute, text for selective download of XML data by mobile clients. It organizes data based on the level of XML document tree and groups XML elements with same XML path prefix to conserve battery power at mobile clients. Experimental results show that proposed method has reduced tune-in time when compared with existing approaches. Hence PSGL approach enhances performance with energy conservation for processing various types of XPath queries.

Modern computing infrastructures should enable realization of converged provisioning and governance operations on virtualized computing, storage and network resources used on behalf of users' workloads. These workloads must have ensured sufficient access to the resources to satisfy required QoS. This requires flexible platforms providing functionality for construction, activation and governance of Runtime Infrastructure which can be realized according to Service Oriented Infrastructure (SOI) paradigm. Implementation of the SOI management framework requires definition of flexible architecture and utilization of advanced software engineering and policy-based techniques. The paper presents an Adaptable SOI Provisioning Platform which supports adaptable SOI provisioning with lightweight virtualization, compliant with the structured process model suitable for construction, activation and governance of IT environments. The requirements, architecture and implementation of the platform are all discussed. Practical usage of the platform is presented on the basis of a complex case study for provisioning JEE middleware on top of the Solaris 10 lightweight virtualization platform.

In most of e-commerce sites evaluation systems are employed to evaluate each user after trades. Normally, seller's evaluation score is shown in e-commerce site and computed based on the score given by buyers. In recent e-commerce sites, the evaluation is based on multiple attributes and buyers give their thinking for each attribute. In this viewpoint, the seller evaluation in e-commerce is collected knowledge from buyers and the synthetic score is computational intelligence because each buyer makes his/her decision whether he/she trades with the seller. In this paper, we discuss the computational intelligence of the evaluation system in e-commerce. Then, to avoid asymmetric and incomplete information in trades, we design a mechanism of the trader evaluation. After that, we present some experiments to show the rate of successful trades in our proposed mechanism. Contributions of this paper are showing the theoretical discussion of e-commerce computational intelligence, design of evaluation mechanism and the effectiveness of the proposed mechanism.

Speed and trend in the software evolution demand the more flexible and available software. For the large-scale software, the availability and automation of current software online upgrade approach is not ideal, which brings the managers much work to maintain all the versions of software. In this paper, we introduce the versioning-based software online upgrade framework (VSOUF) over BitTorrent-like Peer-to-Peer (P2P) Network. The distributed hash table (DHT) layer and version control (VC) layer is the core of this framework. Software clients can carry out initializing and upgrading by the atomic operations of a version control. Borrowing from P2P, we weave the distributed hash table (DHT) algorithm to speed up the download rate of a version control so that this framework will not crash under the failure of a single node. Finally, experiments have showed the high performance and availability of the proposed framework.

The localization of object parts in the component-based object detection is among the main tasks to solve. This paper presents several improvements of the proposed local image descriptor based on Gabor wavelets. Including these descriptors in the desired application is an ambitious challenge if we take into account the high number of parameters. Determining of parameters can be very hard because of their infinite definition range. Defining the filters is done in two stages: a theoretical consideration narrows the domain and the cardinality of parameters; this is followed by adequate experiments to select the most characteristic descriptor for a target image patch. The descriptor is created from a given number of 2D Gabor filters chosen by the GentleBoost learning algorithm. Comparing the proposed descriptor to those found in the state of the art, we can conclude that the selected filters are adaptable to any target object. In contrast to this, the majority of filter-based descriptors have fixed values for the parameters that do not allow to be ductile to the given object. Parameters fine-tuning allows the descriptor to be general, and discriminative at the same time. The effect of the following experiments has been analyzed during the investigation: elimination of redundancy between the weak classifiers, using the LoG interest points in the detection process. Finally, we propose an acceleration algorithm in order to deter- mine the response map faster. By means of the descriptor, the response map is created, which accurately localizes the target object part and can easily be integrated in almost all detection systems.

We propose a method based on combining a constructive and a bounding heuristic to solve the vertex clique covering problem (CCP), where the aim is to partition the vertices of a graph into the smallest number of classes, which induce cliques. Searching for the solution to CCP is highly motivated by analysis of social and other real-world networks, applications in graph mining, as well as by the fact that CCP is one of the classical NP-hard problems. Combining the construction and the bounding heuristic helped us not only to find high-quality clique coverings but also to determine that in the domain of real-world networks, many of the obtained solutions are optimal, while the rest of them are near-optimal. In addition, the method has a polynomial time complexity and shows much promise for its practical use. Experimental results are presented for a fairly representative benchmark of real-world data. Our test graphs include extracts of web-based social networks, including some very large ones, several well-known graphs from network science, as well as coappearance networks of literary works' characters from the DIMACS graph coloring benchmark. We also present results for synthetic pseudorandom graphs structured according to the Erdös-Renyi model and Leighton's model.

We compare GPGPU implementations of two popular models of crowd dynamics. Specifically, we consider a continuous social force model, based on differential equations (molecular dynamics) and a discrete social distances model based on non-homogeneous cellular automata. For comparative purposes both models have been implemented in two versions: on the one hand using GPGPU technology, on the other hand using CPU only. We compare some significant characteristics of each model, for example: performance, memory consumption and issues of visualization. We also propose and test some possibilities for tuning the proposed algorithms for efficient GPU computations.