Inconsistency Values for Logic Programs under Answer Set Semantics

Vortrag von Dominik Rusovac

Veranstaltungsort: online
Beginn: 22. Oktober 2020 um 13:00
Ende: 22. Oktober 2020 um 14:30
Event series: Research Seminar Logic and AI
iCal

The proposition that conflicting pieces of information cannot exist together appears to be too simplistic for developing intelligent systems. Knowledge representation tools such as ASP (Answer Set Programming) rely on non-monotonic reasoning, i.e., programs may contain inconsistent information without collapsing. While measuring inconsistency in monotonic formalisms has been investigated for some time now, my master thesis addresses the issue of measuring inconsistency in ASP in order to provide an approach for analysing and resolving inconsistency in logic-based knowledge representation. Taking nonmonotonicity into account demands for a generalisation of the notion of inconsistency, namely Strong Inconsistency [2]. Aiming to generalise A. Hunter's and S. Konieczny's

approach [1] to non-monotonic formalisms, the key idea is to characterise and use Minimal Strong Inconsistency Values for inducing a game in coalitional form on a knowledge base, and to then use the Shapley Value in order to obtain a solution to this sort of blame game, which reflects the impact of each element on the overall inconsistency in the base. Based on the latter I characterise the Minimal Strong Inconsistency Shapley Strong Inconsistency (MSISSI) value and the MSISSI measure, which provide an approach to describing inconsistency in ASP and allow for repairing inconsistent knowledge bases.

[1] Anthony Hunter and Sébastien Konieczny. "On the measure of conflicts: Shapley inconsistency values". In: Artifcial Intelligence 174.14 (2010), pp. 1007-1026.

[2] Markus Ulbricht, Matthias Thimm, and Gerhard Brewka. "Measuring Strong Inconsistency." In: AAAI. 2018, pp. 1989-1996.

Dominik Rusovac was born in Vienna and studied Philosophy at the University of Vienna. Then he studied Logic at the University of Leipzig, focussing on knowledge representation and formal systems. After graduating in 2019 he has been working since January 2020 as a software developer.

The talk will take around 45 minutes after which there will be the opportunity to ask questions and it will take place online. If there is any interest in attending, please send an e-mail to thomas.feller@tu-dresden.de.