Instructor: Richard Booth (University of Luxembourg and Mahasarakham University)

Time and venue: 13:30-16:00, December 8, Meeting Room, Faculty of Informatics building

Description

Human reasoning often proceeds via the construction of arguments for a given claim. These arguments may typically take the form of a proof, consisting of a number of reasons chained together, for example “I believe A, and the truth of A is a good reason to accept B, so I believe B. Then the truth of B is a good reason to accept C”, and so on. However these proofs may turn out to be defeasible rather than conclusive, leaving them open to attack from counter-arguments. The typical example is “I believe that Tweety is a bird, and birds normally fly, so I believe Tweety flies”, which is attacked by counter-argument “Tweety is a penguin, and penguins are not normally able to fly, so Tweety cannot fly”. Argumentation theory is concerned with the following question:

                Given a collection of arguments, and a given binary “attacks” relation between them, which arguments

                should be considered as “winning”, or acceptable arguments?

In this tutorial I will first present the notion of abstract argumentation framework (AAF) first introduced by Phan Minh Dung (AIT, Pathum Thani) in 1995. In this theory the notion of argument is taken as primitive, i.e., all internal structure of the arguments is abstracted away, and all that matters is to know which arguments attack which other ones. Essentially an AAF may be visualised as a directed graph whose nodes are the arguments and where an edge from A to B indicates that argument A attacks argument B. Then I will go through the main “argumentation semantics” which have been proposed, such as preferred, grounded, stable, and semi-stable. Each of these argumentation semantics provides a “rule” for selecting sets of winning arguments, or extensions, from a given AAF. We will see how each depends on differing intuitions. To illustrate the concepts we will make extensive use of argument labellings, popularised by Martin Caminada, which assign a label in, out, or undec to each node in the argument graph. These labels indicate whether an argument is respectively accepted, rejected or undecided (neither accepted nor rejected). Our aim is to give a basic introduction to a research area which is currently very active in the field of knowledge representation.

About the instructor

Dr Richard Booth is currently a Research Associate at the University of Luxembourg, as well as being affiliated to Mahasarakham University as Adjunct Lecturer. He obtained his PhD at the University of Manchester, UK and has held research positions at the Max Planck Institute for Computer Science, Saarbruecken and Leipzig University (both Germany), Wollongong University and Macquarie University (both Australia). He has published papers in leading international conferences (IJCAI, ECAI, AAAI, KR) and journals (Artificial Intelligence Journal, Journal of Artificial Intelligence Research, Journal of Logic and Computation). His main research interests are logic-based approaches to Artificial Intelligence, especially Belief Revision, Argumentation Theory, Multi-agent Systems and Reasoning about Preferences.