Initial Experiments in Probabilistic Satis abilityMichael

Both satissability problems and probabilistic models are popular in artiicial intelligence and computer science. This paper looks at the rich intersection between these two topics, opening the door for the use of satissability approaches in probabilistic domains. The paper examines a generic probabilistic satissability problem, which can function for probabilistic domains as Sat does for deterministic domains. Extending the analogy with Sat, the paper deenes a Davis-Putnam-Logemann-Loveland-style procedure for solving probabilistic satissability problems, and reports on a preliminary empirical exploration of the complexity of the algorithm for a collection of randomly generated probabilistic problems. The results exhibit the familiar easy-hardest-hard pattern for the diiculty of random Sat formu-lae. Special cases of the probabilistic satissabil-ity problem lie in diierent complexity classes, and one counterintuitive result is that the computational complexity and the empirical complexity of the problems examined do not track each other exactly|problems in the hardest complexity class are not the hardest to solve.