LINQS

STATISTICAL RELATIONAL LEARNING GROUP @ UMD



 

Joint Probabilistic Inference for Causal Structure Discovery

Dhanya Sridhar, Lise Getoor
Uncertainty in Artificial Intelligence (UAI) Workshop on Causation - 2016
Download the publication : sridhar-uai-open-problem-v2.pdf [121Ko]  
Recently, causal structure learning has been cast as a MAX-SAT problem, with d-separation criteria enforcing constraints over assignments to causal edges between variables, similar to constraint-based methods. The constraint-satisfaction viewpoint has been extended to constrained optimization for structure discovery, solved by a linear program. Motivated by these MAX-SAT and optimization-based approaches, we recognize that causal structure discovery can also be viewed as a probabilistic inference problem. We define distributions over model structures and infer the most likely structure given observations, replacing search over structures with optimization. The open problem that we pose is how to incorporate all of this information in a unified and scalable probabilistic framework.

BibTex references

@InProceedings{sridhar:uaiws16,
  author       = "Sridhar, Dhanya and Getoor, Lise",
  title        = "Joint Probabilistic Inference for Causal Structure Discovery",
  booktitle    = "Uncertainty in Artificial Intelligence (UAI) Workshop on Causation",
  year         = "2016",
}

Other publications in the database