Systematic Review of Probabilistic Inference in Bayesian Networks Using Boolean Satisfaction Algorithms and Model Counting
Year: 2026
Type: review
Study design: Systematic Review
Abstract: Probabilistic inference in Bayesian networks remains one of the most challenging problems in artificial intelligence due to its #P-complete nature. This systematic review analyzes the evolution and current state of Boolean Satisfiability (SAT) and model counting (#SAT) based approaches for exact and approximate inference. Following the PRISMA 2020 guidelines, both classical and recent literature are examined to synthesize encoding strategies, knowledge compilation techniques, weighted model counting, and approximation methods with probabilistic guarantees. The findings show that SAT-based inference achieves competitive performance, particularly in networks with low treewidth, and supports efficient multi-query inference after an initial compilation phase. Moreover, hashing-based approximate counting methods provide rigorous (ε, δ) guarantees. Despite these advantages, challenges remain regarding efficient encodings and limited applicability to hybrid or continuous domains. Overall, SAT/#SAT-based methods emerge as a robust complementary paradigm that strengthens probabilistic reasoning in artificial intelligence.
Institution: Universidad Técnica Estatal de Quevedo
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7162527331