[Zurück]

M. Kronegger, A. Pfandler, S. Ordyniak:
"Variable-Deletion Backdoors to Planning";
Poster: Twenty-Ninth AAAI Conference on Artificial Intelligence, Austin, Texas, USA; 25.01.2015 - 30.01.2015; in: "Proceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence, January 25-30, 2015, Austin, Texas, USA", B. Bonet, S. Koenig (Hrg.); AAAI Press, (2015), S. 3305 - 3312.

Backdoors are a powerful tool to obtain efficient algorithms for hard problems. Recently, two new notions of backdoors to planning were introduced. However, for one of the new notions (i.e., variable-deletion) only hardness results are known so far. In this work we improve the situation by defining a new type of variable-deletion backdoors based on the extended causal graph of a planning instance. For this notion of backdoors several fixed-parameter tractable algorithms are identified. Furthermore, we explore the capabilities of polynomial time preprocessing, i.e., we check whether there exists a polynomial kernel. Our results also show the close connection between planning and verification problems such as Vector Addition System with States (VASS).

Projektleitung Reinhard Pichler:
Effiziente, parametrisierte Algorithmen in Künstlicher Intelligenz und logischem Schließen