Catálogo de publicaciones - libros

We present an efficient algorithm for job submissions in Peer-to-Peer (desktop) grids based on limited multicasts. Our approach combines the advantages of two overlay architectures: Chord-like structured networks and unstructured networks with epidemic communication. To predict the multicast properties and to optimize its distribution schedule, we present a probabilistic model of the process of information propagation within the overlay. We show the efficiency and the fault-tolerance of our proposed method and demonstrate the high accuracy of the predictive model.

Building on previous work on how to model contextual information for desktop search and how to implement semantically rich information exchange in social networks, we define a new algorithm, Peer-Sensitive ObjectRank for ranking resources on the desktop. The new algorithm takes into account different trust values for each peer, generalizing previous biasing PageRank algorithms. We investigate in detail, how different assumptions about trust distributions influence the ranking of information received from different peers, and which consequences they have with respect to integration of new resources into one peer’s initial network of resources. We also investigate how assumptions concerning size and quality of a peer’s resource network influence ranking after information exchange, and conclude with directions for further research.

Performance tuning for J2EE application server systems is a complex manual task. This is unfortunately a necessary task in order to achieve optimal performance under dynamic workload environment. In this paper, we present our architecture and approach for implementing autonomic behavior in J2EE application server systems. Our experimental results demonstrate the feasibility and practicality of our architecture and approach in automatic performance tuning of J2EE application server systems.

XML is currently being used as the message syntax for Web services. To enable small mobile devices to use Web services, this XML use must not be too resource-consuming. Due to several measurements indicating otherwise, alternate serialization formats for XML data have been proposed. We present here a format for XML data designed from the ground up for the mobile environment. The format is simple, yet gives acceptable document sizes and is efficiently processable. An automaton-based approach gives further improvements when full or partial schema information is available. We provide performance measurements verifying these claims and also consider some issues arising from the use of an alternate XML serialization format.

The ability to store information contained in XML documents for future references becomes a very important issue as the number of applications which use and exchange data in XML format is growing continuously. Moreover, the contents of XML documents are dynamic and they change across time. However, storing all document versions in an XML data warehouse would introduce a high level of redundancy. Nevertheless the ability to store XML documents together with their different versions across time is often required. Our paper proposes a novel approach for storing changes of dynamic XML documents in time with less overhead so earlier versions can be easily queried. We show how our proposed consolidated delta is built, with steps and rules of the algorithm involved and we demonstrate the efficiency of the versioning approach in terms of storage and retrieval using some test data.

The Web is a distributed network of information sources where the individual sources are autonomously created and maintained. Consequently, syntactic and semantic heterogeneity of data among sources abound. Most of the current data cleaning solutions assume that the data values referencing the same object bear some textual similarity. However, this assumption is often violated in practice. “Two-door front wheel drive” can be represented as “2DR-FWD” or “R2FD”, or even as “CAR TYPE 3” in different data sources. To address this problem, we propose a novel two-step automated technique that exploits statistical dependency structures among objects which is invariant to the tokens representing the objects. The algorithm achieved a high accuracy in our empirical study, suggesting that it can be a useful addition to the existing information integration techniques.

Since the evaluation of XPath expressions is highly dependent upon their size and navigational structures that include ancestor-descendant relationships (“//”) and wildcard steps (“/∗”), we introduce a novel and complementary approach to optimizing XPath queries by rewriting and minimizing such structural occurrences. This rewriting approach depends upon the existence of a statistical schema, which we derive from a set of pre-processed XML documents. However, an imprecision in the schema extraction may lead to a loss of accuracy in the results. Through experimentation and analysis, we validate the scalability and efficiency of our approach.

In this paper, we propose a scalable performance management tool for Wide Area Applications. Our objective is to scalably identify non-random associations between pairs of individual Latency Profiles (iLPs) (i.e., latency distributions experienced by clients when connecting to a server) and exploit them in latency prediction. Our approach utilizes Relevance Networks (RNs) to manage tens of thousands of iLPs . Non-random associations between iLPs can be identified by topology-independent measures such as correlation and mutual information. We demonstrate that these non-random associations do indeed have a significant impact in improving the error of latency prediction.

In this work, we propose an effective online face recognition system over the error-prone Internet. The proposed system uses multiple JPEG-compressed images to improve the recognition rate, and image compression and resizing to reduce the transmission delay and the processing delay. Furthermore, the robustness is improved by effective packetization of compressed image data. First of all, we examine the face recognition rate in terms of quantization parameter and image size, and then implement the effective multiple-image-based face recognition system. Finally, experimental results are provided to show the performance comparison with existing algorithms over the error-prone Internet environment.

This paper discusses the concept of information market. The authors of this paper have been involved in several aspects of information retrieval research. In continuing this research tradition we now take a wider perspective on this field and re-position it as a market where demand for information meets supply for information. The paper starts by exploring the notion of a market in general and is followed by a specialization of these considerations in the information market, where we will also position some of the existing work.