Satu Eloranta defends her PhD thesis on June 27th, 2013 on Dynamic Aspects of Knowledge Bases

Ph.Lic. Satu Eloranta will defend her doctoral thesis Dynamic Aspects of Knowledge Bases on Thursday 27th of June 2013 at noon in the University of Helsinki Exactum Building, Gustaf Hällströmin katu 2b, Linus Torvalds Auditorium B123, 1st floor. Her opponent is Professor Antti Valmari (Tampere University of Technology) and custos Professor Esko Ukkonen (University of Helsinki). The defense will be held in Finnish.

Dynamic Aspects of Knowledge Bases

A knowledge base is considered a system that is told information about an external world and that answers questions about this world. Our goal here is to outline knowledge bases that involve both knowledge and beliefs. In previous studies, various kinds of belief change have been studied in isolation, but we want to tie them together. We aim at knowledge bases that could carry the epistemic states of agents, that is, the knowledge and the beliefs that an agent has at any one moment in time.

The difference between knowledge and belief is that while knowledge increases monotonically with time, beliefs may at some later point in time turn out to be false. Beliefs may change for various reasons: in belief revision, beliefs are changed when receiving new information about a world that has not changed, while in belief update a change in the world is to be recorded. Different types of change call for different treatments. In belief-change studies, various change types have been characterized by rationality criteria set on each type. The main principles in these criteria are maintaining consistency of beliefs and minimality of change.

When dealing with belief change, our approach is to take knowledge as an integrity constraint that should always hold, and we describe how the rationality criteria should be modified accordingly. In our refined rationality criteria, beliefs that are inconsistent with the knowledge of the knowledge base will never be allowed to enter into the knowledge base.

In the rationality criteria, a common assumption is that the most recent information is the most reliable, and it has therefore been prioritized over the old beliefs. However, this may not be the case in all circumstances. In order to complete the collection of belief-change types, we propose a new, commutative type of change for entering competing evidence into the knowledge base.

The representation theorems that have been given for belief revision indicate that belief revision involves an ordering of disbelief on possible alternative situations, or equivalently, an epistemic entrenchment on logical formulas. A formula less entrenched is more easily given up when eliminating inconsistancies. In view of the changes in the rationality criteria, we also refine the representation theorems.

We introduce two finite representations for knowledge bases, one with a finite ordered set of propositional formulas that are satisfiable but pairwise inconsistent with each other, and the other with a finite list of pairwise inconsistent propositional formulas. Both representations involve dynamic orderings of disbelief that have arisen out of the previous change operations.

We show that for the knowledge base to satisfy the rationality criteria given for belief revision, the dynamic ordering of disbelief in the knowledge base is vital. The representations and the operators that we introduce in this thesis demonstrate how this ordering of disbelief could be dealt with in various operations.

Availability of the dissertation

An electronic version of the doctoral dissertation is available on the e-thesis site of the University of Helsinki at http://urn.fi/URN:ISBN:978-952-10-8990-9.

Printed copies are available on request from Satu Eloranta: satu.eloranta@cs.helsinki.fi.

11.07.2013 - 12:10 Pirjo Moen
19.06.2013 - 09:36 Pirjo Moen