Reasoning with preference in only knowing logic with confidence levels

Abstract

The purpose of this thesis is to present an autoepistemic approach to preference reasoning. The method controls the generation of extensions by first linking doxastic representations in a hierarchy, and then by ensuring the maintenance of integrity between the levels in the hierarchy using so-called integrity constraints. This approach is based on an encoding of ordered default theories into an ''only knowing'' logic with confidence levels, and is the result of studying several principles for default logic approaches to preference, as well as a critique of the hierarchic autoepistemic logic of Konolige with respect to these same principles. We show that our approach is the first autoepistemic formalism that handles preference according to all the principles mentioned above. Another main result of our work is that the generation of preferred extensions may be carried out entirely at the object level. Thus, as opposed to approaches to preference using default logic, our approach provides a constructive method for determining extensions of ordered default theories.