Catálogo de publicaciones - libros

Much of modern scientific research depends on databases, but do we need anything more than conventional database technology to support scientific data? One of the reasons for the development of the Grid is the sheer size of the datasets involved. This has introduced new problems for distributed data, distributed scientific programming, and the combination of the two. However there are other, equally important issues which demand new database technology. In this talk I want raise some of them.

Annotation of existing data now provides a new form of communication between scientists, but conventional database technology provides little support for attaching annotations. I shall show why new models of both data and query languages are needed.

Closely related to annotation is provenance — knowing where your data has come from. This is now a real problem in bioinformatics with literally hundreds of databases, most of which are derived from others by a process of transformation, correction and annotation.

Preserving past states of a database — archiving — is also important for verifying the basis of scientific research, yet few published scientific databases do a good job of archiving. Past “editions” of the database get lost. I shall describe a system that allows frequent archiving and efficient retrieval with remarkably little space overhead.

Finally, what do scientific databases have to do with multimedia information systems? Ostensibly nothing. However presentation of data has given us some clues about how to approach some of the problems above.

Nowadays content and services are available at different sources and places, thus a user can be seen as an integral part of numerous applications in which he/she interacts with service providers, product sellers, governmental organisations, friends and colleagues.

Information access personalization can be defined as any set of actions that can tailor information to a particular user or set of users. To achieve effective personalization, single users and organizations must rely on all available information, including: user profile, channel peculiarities, users current interests and typical behaviour, source content, source structure, as well as domain knowledge. In addition, efficient and intelligent techniques are needed to effectively use the discovered knowledge to enhance the users’ experience.

These techniques must address important challenges emanating from the size and the heterogeneous nature of the data itself, as well as the dynamic nature of user interactions with data sources. These challenges include scalability of the personalization solutions in the process of data integration, and successful integration of techniques from knowledge representation, semantic web, information retrieval and filtering, databases, user modelling and human-computer interaction.

Our approach addresses these issues with particular attention to the process of data tailoring, which consists of the exploitation of knowledge about , the and the , altogether called , to the end of reducing the amount of information imported on the mobile device. Tailoring is needed because of two main reasons: one is the need to keep information manageable, in order for the user not to be confused by too much noise; the second reason is the frequent case that the mobile device be a small one, like a palm computer or a cellular phone, in which condition only the most precious information must be kept on board.

We consider open, networked, peer-based systems, according to paradigms where there is no previous knowledge and relationship among the parties, which may be mobile as well as fixed devices. The interaction among such devices is transient, since it is subject to network and device availability: indeed the nature of these devices and the variety of ambient strongly affect the cooperation methods and techniques. Semantic based caching techniques are exploited to cope with the above mentioned network availability problems, always allowing the single party to retain the appropriate portion of needed data, while other fragments, stored at different peers, can be queried only when a connection is available. The mobile and dynamic context where the devices cooperate determines the fraction of data located on board, which, due to the limited amount of memory available, must be refreshed according to semantic context-based criteria. On the other hand, power aware data access techniques must be employed to manage the problem of limited battery life.

Consider the example of a semantic community formed to enable scientific collaboration in the medicine context; here, different resource structures and meanings are provided by the community peers. Besides selecting and semantically integrating the most appropriate resources provided by the various peers, it is the special goal of such techniques to obey the constraints imposed by the device context. For instance, during a home visit, a doctor in need of information about the symptoms of a rare disease can search, through his/her PDA or cell phone, clinical databases and research reports on the web looking for assistance in his/her diagnosis. Obviously the doctor must not be disoriented by the different formats of the retrieved documents, by the possible lexical discrepancies in their contents and by information which would be useless in the specific environment (tropical diseases symptoms in an Eskimo patient), but the most valuable information must be presented in the most suitable form for the operational context and the available device.

In this work we describe the driving the selection of the device-residing portion of data; for the detection of the context provide the different perspectives the mobile device is viewed from, and are used to set out its . The identified dimensions and their current values drive the choice of the information to be kept on the mobile device, to be actually selected at run time. In order to formalize and then automatically obtain this view (the device ambient), we model the dimensions as a hierarchical, DAG-shaped, structure which allows us to consider an ontological specification of each considered concept, and to model semantic constraints between different dimensions as well. The dimension DAG contributes to the automatic selection and interpretation of the shared resources to be imported to the device.

Moreover, when interesting concepts are found and selected from the available data sources, it is also useful to collect on the device other concepts which are possibly related to them. However, since devices may have a limited memory, dynamic conflict resolution strategies must be devised, possibly based on various notions of semantic nearness, to decide which data must be retained and which can be discarded.

The development of technology generates huge amounts of non-textual information, such as images. An efficient image annotation and retrieval system is highly desired. Clustering algorithms make it possible to represent visual features of images with finite symbols. Based on this, many statistical models, which analyze correspondence between visual features and words and discover hidden semantics, have been published. These models improve the annotation and retrieval of large image databases. However, image data usually have a large number of dimensions. Traditional clustering algorithms assign equal weights to these dimensions, and become confounded in the process of dealing with these dimensions.

In this tutorial, first, we will present current state of the art and its shortcomings. We will present some classical models (e.g., translation model (TM), cross-media relevance model etc.). Second, we will present weighted feature selection algorithm as a solution to the existing problem. For a given cluster, we determine relevant features based on histogram analysis and assign greater weight to relevant features as compared to less relevant features. Third, we will exploit spatial correlation to disambiguate visual features, and spatial relationship will be constructed by spatial association rule mining. Fourth, we will present the continuous relevance model and multiple Bernoulli model for avoiding clustering. We will present mechanisms to link visual tokens with keywords based on these models. Fifth, we will present mechanisms to improve accuracy of classical model, TM by exploiting the WordNet knowledge-base. Sixth, we will present a framework to model semantic visual concept in video/images by fusing multiple evidence with the usage of an ontology. Seventh, we will show that weighted feature selection is better than traditional ones (TM) for automatic image annotation and retrieval. Finally, we will discuss open problems and future directions in the domain of image and video.

Most face based biometric systems and the underlying recognition algorithms are often more suited for verification (one-to-one comparison) instead of identification (one-to-many comparison) purposes. This is even more true in case of large face database, as the computational cost of an accurate comparison between the query and a gallery of many thousands of individuals could be too high for practical applications. In this paper we present a 3D based face recognition method which relies on normal image to represent and compare face geometry. It features fast comparison time and good robustness to a wide range of expressive variations thanks to an expression weighting mask, automatically generated for each enrolled subject. To better address one-to-many recognition applications, the proposed approach is improved via DFT based indexing of face descriptors and --tree based spatial access to clusters of similar faces. We include experimental results showing the effectiveness of the presented method in terms of recognition accuracy and the improvements in one-to-many recognition time achieved thanks to indexing and retrieval techniques applied to a large parametric 3D face database.

Information retrieval is moving beyond the stage where users simply type one or more keywords and retrieve a ranked list of documents. In such a scenario users have to go through the returned documents in order to find what they are actually looking for. More often they would like to get targeted answers to their queries without extraneous information, even if their requirements are not well specified. In this paper we propose an approach for designing a web retrieval system able to find the desired information through several interactions with the users. The proposed approach allows to overcome the problems deriving from ambiguous or too vague queries, using semantic search and topic detection techniques. The results of the very first experiments on a prototype system are also reported.

The importance of web discussion boards is growing with the interest of sharing knowledge and doubts with colleagues in a working/studying environment. The challenge is to organize the structure of discussion boards, to make the navigation easier, and to effectively extract relevant information. Message hierarchies in web discussion boards, manually organised by users participating into the discussion, might grow uncontrolled, thus making navigation more and more difficult for users. The goal of this paper is to develop a technique to organise messages in a message board, by automatically classifying and annotating pairs of postings to guide users through discussion segments relevant to their navigational goals.

The traditional multidimensional models have a static structure where members of dimensions are computed in a unique way. However, data (particularly multimedia data) is often characterized by descriptors that can be obtained by various computation modes. We define these computation modes as "functional versions" of the descriptors. We propose a Functional Multiversion Multidimensional Model ("F2M model") by integrating the concept of "version of dimension". This concept defines dimensions with members computed according to various functional versions. In order to allow the user to choose the best representation of data, this new approach integrates a choice of computation modes of these members into the model. We implement a multimedia data warehouse in the medical field by integrating the multimedia data of a therapeutic study into a multidimensional model. We formally define a conceptual model and we present a prototype for this study.

In this paper we introduce a new indexing approach to representing multimedia object classes generated by the Expectation Maximization clustering algorithm in a balanced and dynamic tree structure. To this aim the EM algorithm has been modified in order to obtain at each step of its recursive application balanced clusters. In this manner our tree provides a simple and practical solution to index clustered data and support efficient retrieval of the nearest neighbors in high dimensional object spaces.

Music description is nowadays considered an important matter in Information and Communication Technology. The encoding formats commonly accepted and employed are often characterized by a partial view of the whole problem: they describe music data or metadata for score, audio tracks, computer performances of music pieces, but they seldom encode all these aspects together. In this paper, we present the MX formalism that aims to address this limitation of the existing formats, by providing a Semantic Web compatible representation of music information in terms of structural and semantic features, by means of XML and OWL.

The icon language interface is designed to provide the people with language disabilities with more smooth and convenient communication environment than the traditional keyboard-based input system. In order for that, vocabulary commanding tendencies and characteristics of proto-corpus built upon the frequently-used conversation sections will be analyzed, and the Korean language’s vocabulary and their meanings applied to the icons will be retrieved by the use of morpheme, phrase and semantic analysis techniques. The icon types to be intuitively recognized and communicable by users are selected, and they are matched with the retrieved Korean vocabulary and meanings. In order to create a relevant situation to communicate from the relations between neighboring icons, the icon language is formed through the each definition of icon language’s vocabulary, grammar rules, parts of speech and semantic system.