Complexity of Projection with Stochastic Actions in a Probabilistic Description Logic

Talk by Benjamin Zarrieß

Location: APB 3027
Start: 20. September 2018 at 1:00 pm
End: 20. September 2018 at 2:30 pm
Research group: Computational Logic
Research group: Automata Theory
Research group: Knowledge Representation and Reasoning
Research group: Knowledge-Based Systems
Research group: Verification and formal quantitative Analysis
Event series: KBS Seminar
iCal

Integrating probabilistic notions of uncertainty into languages for reasoning about actions is a popular approach to adequately deal for instance with possibly fallible acting and sensing. In this talk, an action language extended with quantitative notions of uncertainty is considered. In our setting, the initial beliefs of an agent are represented as a probabilistic knowledge base with axioms formulated in the Description Logic ALCO. Action descriptions describe the possibly context-sensitive and nondeterministic effects of actions and provide likelihood distributions over the different possible outcomes. A decidability result for the probabilistic projection problem, which is the basic reasoning task needed for predicting the outcome of action sequences, is presented. Furthermore, I will talk about the problem of how the nondeterminism in the action model affects the complexity of the projection problem.