Catálogo de publicaciones - libros

Logic of plausible reasoning (LPR) is a formalism which is based on human inference patterns. In the paper the LPR is defined as a labeled deductive system. Knowledge base consists of labeled formulas representing object-attribute-value triples, implications, hierarchies, dependencies and similarities between objects. Labels are used to represent plausible parameters. In the paper LPR basic semantics is defined and the proof system is proved to be correct. Finally, several examples of inference pattern application are presented.

In this paper, we propose a logic program based formulation that supports delegatable authorizations, where negation as failure, classical negation and rules inheritance are allowable. A conflict resolution policy has been developed in our approach that can be used to support the controlled delegation and exception. In our framework, authorization rules are specified in a Delegatable Authorization Program (DAP) which is an extended logic program associated with different types of partial orderings on the domain, and these orderings specify various inheritance relationships among subjects, objects and access rights in the domain. The semantics of a DAP is defined based on the well-known stable model and the conflict resolution is achieved in the process of model generation for the underlying DAP. Our framework provides users a feasible way to express complex security policies.

This paper deals with the querying of databases containing ill-known attribute values represented by possibility distributions. It investigates a query language with two operators: selection and a foreign key join operator which allows to express queries involving several relations at a time. The key idea is to impose a strong connection between the relations resulting from the operators and their interpretation in terms of more or less possible worlds. From a computational point of view, an interesting property of the queries considered is that they can be evaluated in a “compact” way, i.e., they do not require explicit handling of the possible worlds attached to the possibilistic database.

OLAP operations have been widely accepted as a suitable method for decision support by data analysis. Among others, roll-up and drill-down are practically implemented by using database operations. However, we cannot define them as inverses of each other since they assume how to manage materialized views. In this work, we model these operators in the framework of meta objects, and extend relational algebra by introducing meta operators group and apply. Then we show two OLAP operations can be managed within the framework of (new) database operations.

Query processing plays a fundamental role in current information systems that need to access independent and heterogeneous databases. This paper presents a new approach to querying heterogeneous databases that maps the semantics of query objects onto database schemas. The sematics is captured by the definitions of classes in an ontology, and a similarity function identifies not only equivalent but also semantically similar classes associated with a user’s request. These similar classes are then mapped onto a database schema, which is compared with schemas of heterogeneous databases to obtain entities in the databases that answer the query.

Personalization has been recently one of the most important features of intelligent information retrieval. An intelligent system should store information about user interests and utilize this information to deliver to the user documents he really needs. In such a system the information needs of a user should be represented by means of so called a user profile. User profiles, in other hand, should be used together with queries to sort retrieved information in such order that is adequate to user preferences. In this paper a vector-based information system model is presented, in which the user information needs and preferences (profiles) are defined and the methods for updating user profiles and automatic learning about user preferences are worked out.

Regular language learning from positive examples alone is infeasible. Subclasses of regular languages, though, can be inferred from positive examples only. The most common approach for learning such is the specific-to-general technique of merging together either states of an initial finite state automaton or nonterminals in a regular grammar until convergence.

In this paper we seek to unify some language learning approaches under the general-to-specific learning scheme. In automata terms it is implemented by refining the partition of the states of the automaton starting from one block until desired decomposition is obtained; i.e., until all blocks in the partition are uniform according to the predicate determining the properties required from the language.

We develop a series of learning algorithms for well-known classes of regular languages as instantiations of the same master algorithm. Through block decomposition we are able to describe in the same scheme, e.g., the learning by rote approach of minimizing the number of states in the automaton and inference of -reversible languages.

Under the worst-case analysis partition-refinement is less efficient than alternative approaches. However, for many cases it turns out more efficient in practice. Moreover, it ensures the inference of the canonical automaton, whereas the state-merging approach will leave excessive states to the final automaton without a separate minimization step.

In this paper, we propose a sound approach and an algorithm for computing a condensed representation of either full or iceberg datacubes. A novel characterization of datacubes based on dimensional-measurable partitions is introduced. From such partitions, iceberg cuboids are achieved by using constrained product linearly in the number of tuples. Moreover, our datacube characterization provides a loss-less condensed representation specially suitable when considering the storage explosion problem and the I/O cost. We show that our algorithm turns out to an operational solution more efficient than competive proposals. It enforces a lecticwise and recursive traverse of the dimension set lattice and takes into account the critical problem of memory limitation. Our experimental results shows that is a promising candidate for scalable computation.

We propose the first paradigm that brings Feature Subset Selection to the realm of ILP, in a setting where examples are expressed as non-recursive Datalog Horn clauses. The main idea is to approximate the original relational problem by a multi-instance attribute-value problem, and to perform Feature Subset Selection on that modified representation, suitable for the task. The method acts as a filter: it preprocesses the relational data, prior to model building, and produces relational examples with empirically irrelevant literals removed. An implementation of the paradigm is proposed and successfully applied to the biochemical mutagenesis domain.

This paper tackles the problem of knowledge discovery in text collections and the dynamic display of the discovered knowledge. We claim that these two problems are deeply interleaved, and should be considered together. The contribution of this paper is fourfold: (1) description of the properties needed for a high level representation of concept relations in text (2) a stochastic measure for a fast evaluation of dependencies between concepts (3) a visualization algorithm to display dynamic structures and (4) a deep integration of discovery and knowledge visualization, i.e. the placement of nodes and edges automatically guides the discovery of knowledge to be displayed. The resulting program has been tested using two specific data sets based on the specific domains of molecular biology and WWW howtos.