Learning sparse causal models is not NP-hard

Open Access
Authors
Publication date 2013
Host editors
  • A. Nicholson
  • P. Smyth
Book title Uncertainty in artificial intelligence: proceedings of the twenty-ninth conference (2013): July 12-14, 2013, Bellevue, Washington, United States
Event Conference on Uncertainty in Artificial Intelligence; 29 (Bellevue, Wash.)
Pages (from-to) 172-181
Publisher Corvallis, Oregon: AUAI Press
Organisations
  • Faculty of Science (FNWI) - Informatics Institute (IVI)
Abstract
This paper shows that causal model discovery is not an NP-hard problem, in the sense that for sparse graphs bounded by node degree k the sound and complete causal model can be obtained in worst case order N2(k+2) independence tests, even when latent variables and selection bias may be present. We present a modification of the well-known FCI algorithm that implements the method for
an independence oracle, and suggest improvements for sample/real-world data versions. It does not contradict any known hardness results, and does not solve an NP-hard problem: it just proves that sparse causal discovery is perhaps more complicated, but not as hard as learning minimal Bayesian networks.
Document type Conference contribution
Language English
Published at http://auai.org/uai2013/prints/papers/121.pdf
Downloads
Permalink to this page
Back