Ho Pun Lam
On the Derivability of Defeasible Logic

In this thesis we focus upon the problem of derivability of rules from Defeasible Logic. Derivability of rules, as defined here, comprises the concept of an extension from a defeasible (Nute 1994) as well as the classical notion of derivability of rules in logic. The idea of localness of reasoning, reasoning with a limited access to rules, is realized by the concept of relative derivability (Gomolinska 2002.

Starting with the derivability of rules, we next touch upon the questions of the activation of rules and (in)consistency of rules in defeasible logic. We will discuss the problems of computing extensions of defeasible logic under different intuitions, with particular interest in ambiguity blocking, ambiguity propagation, well-founded semantics, and their combinations, and will present algorithms to these problems. In addition, we will also discuss the deficiency of the current inference process and will present a new theorem and an algorithm to enhance the process. Matters related to the implementations issues of the algorithms and experimental results will also be discussed in this thesis.

Next we will discuss the problem of ambient intelligence: the imperfect nature of context, the dynamic nature of ambient environment, and special characteristics of the devices involved, which imposed new challenges in the field of distributed Artificial Intelligence. This thesis proposes a solution based on Multi-Context paradigm with non-monotonic features. We will model ambient agents as logic-based entities and present a framework based on the argumentation semantics of defeasible logic. We also extends the semantics of modal defeasible logic in handling violations and preferences, and can derive the conclusion with the avoidance of transformations. We also present an operational model in the form of distributed reasoning algorithm for query processing with iterative revision based on speculative computation. Issues such as inconsistent arguments handling and agents context preferences will also be discussed.

Last but not least, a prototypical implementation showcasing the wealth of our approach will also be presented at the end of the thesis.

(pdf, BiBTeX)