# Relazioni invitate

## Relazioni invitate

- Terrance Swift, Stony Brook University.
How Tabling Solves Real Problems

*24 Giugno 2009**Abstract*: In its early days, Tabled Logic Programming (TLP) was primarily used on definite programs to ensure the termination and efficient execution of queries. Since then, a number of sophisticated tabling mechanisms have been robustly implemented in Prolog systems such as XSB, YAP and Bin-Prolog. Beyond its most common uses for definite programs, TLP can be used to implement the 3-valued Well-Founded Semantics, which is of interest both in itself and as a means to interface Prolog programs with ASP solvers. Implementations of tabling have also been extended to interact with constraints; to use call subsumption, which makes model generation more efficient; and to support answer subsumption, which can be used for quantitative and constraint-based reasoning.

Furthermore, there are now parallel and multi-threaded implementations of tabling. While some of these features have been recently developed, many have been used in a number of research and commercial applications. This talk discusses how some of these approaches can be used to solve applications in verification, the semantic web and machine learning. - Manfred Jaeger, Aalborg Universitet
Probabilistic Logic Models: Expressivity and Inference

*25 Giugno 2009**Abstract*: The integration of logic and probability has been pursued at least since the mid-19th century, when George Boole developed the first propositional probability logic. This logic, like most of its successors, is essentially a multi-valued logic with probabilities replacing binary truth values. For practical knowledge representation and reasoning tasks these logics have met with only limited success. Main obstacles for their applicability are their lack of truth-functionality, and their limited support for reasoning with stochastic independence information.

A different approach to combining probability and logic has arisen out of Artificial Intelligence and Machine Learning during the last 15 years. Variously called 'Statistical Relational Learning', 'Probabilistic Logic Learning (PLL)', or 'Probabilistic Inductive Logic Programming', this approach uses logic-based representation languages to specify concrete probabilistic models, rather than probabilistic-logic theories.

The large number of different PLL frameworks that have been proposed has led to a need for better understanding their relationships in terms of semantics, expressivity, complexity, and learnability. In this talk I will present a uniform semantic framework for PLL languages, based on which expressivity can be compared. I will introduce the two concrete representation languages 'Relational Bayesian Networks' and 'Markov Logic Networks', and apply the given framework to show that RBNs are at least as expressive as MLNs. Also in this talk I will outline some challenging inference tasks for PLL frameworks which are closer in spirit to logical entailment than conventional PLL inference, but which have not been much considered so far.

## Tutorial

Massimiliano Giacomin, Università degli Studi di Brescia

Abstract argumentation and semantics: an introduction

*26 Giugno 2009*

*Abstract*: Argumentation Theory is a framework for practical and uncertain reasoning viewed as a process of arguments production and evaluation. Arguments are entities including a supported conclusion and a set of premises that represent reasons to believe the conclusion itself. An argument can represent, for example, the deduction of a conclusion based on a set of logical rules, or on the basis of empirical knowledge, or it can model a form of abductive reasoning. More generally, the conclusion of an argument can be supported by incomplete and uncertain knowledge, with the result that different arguments may be in conflict.

In order to provide a general unifying view able to encompass most of the existing approaches to argumentation, an abstract framework has been introduced by Dung, which leaves unspecified the origin and the structure of arguments and models the interaction between them simply as a binary relation indicating that an argument attacks another one. This way, it is possible to focus exclusively on the definition of an argumentation semantics, i.e. a method to determine from a given set of arguments the subset of acceptable ones, namely those arguments that can be believed given the conflict relationships involving the whole set.