Catálogo de publicaciones - libros

Knowledge and Information distribution, which is one of the main processes in Knowledge Management, is greatly affected by explicit relations, as well as by implicit relations like . Making decisions about whether to deliver or not a specific piece of information to users on the basis of a rationally justified procedure under potentially conflicting policies for power and trust relations is indeed a challenging problem. In this paper we model power relations, as well as delegation and trust, in terms of an argumentation formalism, in such a way that a dialectical process works as a decision core, which is used in combination with the existing knowledge and an information distribution system. A detailed example is presented and an implementation reported.

This paper presents the pertinence of the use of the Answer Set Programming (ASP) formalism for developing a computational model of a software agent for Computer Supported Collaborative Learning (CSCL) environments. This analytic model is based on a representation of for agent’s beliefs about the learner and the domain, together with the corresponding inference system with the appropriate rules to derive new beliefs about the capabilities of the learner, and its use in order to support effective collaboration and maintain learning possibilities for the group members. The model provides a representation of the structural knowledge frontier and the social knowledge frontier of the learner, which are the components for the definition of the learner’s zone of proximal development (zpd). Based on the zpd of its learner the agent can propose her a learning task and maintain the zpd for the learner in the group. The complete code of the model is presented in the declarative language of DLV, a logic programming language for implementing ASP models.

It is shown in this paper a way of representing deductive systems using the situation calculus. The situation calculus is a family of first order languages with induction that allows the specification of evolving worlds and reasoning about them and has found a number of applications in AI. A method for the representation of formulae and of proofs is presented in which the induction axiom on states is used to represent structural induction on formulae and proofs. This paper’s formalizations are relevant for the purpose of meta reasoning and of automated or manual deduction in the context of situation calculus specifications. An example proof is given for the fact that no deductive system is complete for arbitrary situation calculus specifications (an expectable result).

#SAT is a complex problem equivalent to calculate the cardinalities of the null sets of conjunctive forms consisting of clauses with an uniform length. Each such null set is the union of linear varieties of uniform dimension in the hypercube. Here we study the class of sets in the hypercube that can be realized as such null sets. We look toward to characterize their cardinalities and the number of ways that they can be expressed as unions of linear varieties of uniform dimension. Using combinatorial and graph theory argumentations, we give such characterizations for very extremal values of , either when it is very small or close to the hypercube dimension, and of the number of clauses appearing in an instance, either of value 2, or big enough to get a contradiction.

In the Belief Change domain, Katsuno and Mendelzon have proposed a set of postulates that should be satisfied by update operators. In 1989, Forbus semantically defined an update operator that satisfies these postulates. In order to calculate the resulting belief base all models of the relevant belief bases must be known. This paper proposes to use the and normal forms to represent these bases. Using this representation, a syntactical and computationally cheaper version of Forbus belief update operator is defined and a new minimal distance is proposed. We claim that this minimal distance ensures a better commitment between the minimal change criterion and the belief update definition.

Based on the idea that the cut sets of fuzzy sets are indeed crisp, but facilitate a normative theory for formalizing fuzzy set theory, this paper introduces cut sets of the fuzzy concepts and fuzzy roles as atomic concepts and atomic roles to build , a new fuzzy extension of . This paper gives the definition of syntax, semantics and knowledge base of and discusses the comparison among and other fuzzy extensions of . In addition, this paper defines the acyclic TBox form of , presents sound and complete algorithms for reasoning tasks w.r.t acyclic TBox, and proves the complexity of them is PSPACE-complete.

In constraint programming, a priori choices statically determine strategies that are crucial for resolution performances. However, the effect of strategies is generally unpredictable. We propose to dynamically change strategies showing bad performances. When this is not enough to improve resolution, we introduce some meta-backtracks. Our goal is to get good performances without the know-how of experts. Some first experimental results show the effectiveness of our approach.

A major technique to address state explosion problem in model checking is abstraction. Predicate abstraction has been applied successfully to large software and now to hardware descriptions, such as Verilog. This paper evaluates the state-of-the-art AI techniques—constraint logic programming (CLP)—to improve the performance of predication abstraction of hardware designs, and compared it with the SAT-based predicate abstraction techniques. With CLP based techniques, we can model various constraints, such as bit, bit-vector and integer, in a uniform framework; we can also model the word-level constraints without flatting them into bit-level constraints as SAT-based method does. With these advantages, the computation of abstraction system can be more efficient than SAT-based techniques. We have implemented this method, and the experimental results have shown the promising improvements on the performance of predicate abstraction of hardware designs.

Grouping problems arise in many applications, and the aim is to partition a set of items, into a collection of mutually disjoint subsets or groups. The objective of grouping is to optimize a cost function such as to minimize the number of groups. Problems in this category may come from many different domains such as graph coloring, bin packing, cutting stock, and scheduling. This investigation is related in particular to scheduling transportation events, modeled as a grouping problem, and with the objective to minimize the number of vehicles used and satisfying the customer demand. There is a set of events to be scheduled (items) into a set of vehicles (groups). Of course, there are constraints that forbid assigning all events to a single vehicle. Two different techniques are used in this work to tackle the problem: Grouping Genetic Algorithms and an algorithm based on the heuristic DJD widely used for solving bin packing problems. Both methods were adapted to the problem and compared to each other using a set of randomly generated problem instances designed to comply with real situations.

We introduce a simple approach to finding approximate solutions to combinatorial problems. This approach called the Radial Search (RS) uses the concept of rings which define the location and size of search areas around current good solutions. It iteratively modifies the radii of these rings, and generates new centres, in order to cover the search space. A concentration step corresponds to choosing a solution as the centre of a new ring. An expansion step corresponds to the exploration around a given centre by increasing and reducing the radius of the ring until a better solution than the centre is found. This dynamic process of concentration and expansion of the search is repeated until a stopping condition is met.

A detailed description of RS, a discussion of its similarities and differences with current known methods, and its performance on TSP and QAP problems are given.