Causal models have no complete axiomatic characterization

2008/04/15 by Li, Sanjiang
#Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Logic in Computer Science (cs.LO)

paper · doi:10.48550/arxiv.0804.2401

Abstract

Markov networks and Bayesian networks are effective graphic representations of the dependencies embedded in probabilistic models. It is well known that independencies captured by Markov networks (called graph-isomorphs) have a finite axiomatic characterization. This paper, however, shows that independencies captured by Bayesian networks (called causal models) have no axiomatization by using even countably many Horn or disjunctive clauses. This is because a sub-independency model of a causal model may be not causal, while graph-isomorphs are closed under sub-models.

Related

• Causal Modeling
• A Bayesian Approach to Learning Causal Networks
• Incorporating Causal Prior Knowledge as Path-Constraints in Bayesian Networks and Maximal Ancestral Graphs
• Causal Theories: A Categorical Perspective on Bayesian Networks
• On the Semi-Markov Equivalence of Causal Models