A SAT-based approach to rigorous verification of Bayesian networks

TL;DR

This paper presents a novel SAT-based verification framework for Bayesian networks, enabling formal property verification to enhance safety and interpretability in critical applications.

Contribution

It introduces a two-step compilation scheme and formal verification queries specifically designed for Bayesian networks, filling a gap in their formal analysis methods.

Findings

Efficient verification of Bayesian network properties demonstrated.

Framework applicable to real-world safety-critical scenarios.

Benchmark results show practical utility and scalability.

Abstract

Recent advancements in machine learning have accelerated its widespread adoption across various real-world applications. However, in safety-critical domains, the deployment of machine learning models is riddled with challenges due to their complexity, lack of interpretability, and absence of formal guarantees regarding their behavior. In this paper, we introduce a verification framework tailored for Bayesian networks, designed to address these drawbacks. Our framework comprises two key components: (1) a two-step compilation and encoding scheme that translates Bayesian networks into Boolean logic literals, and (2) formal verification queries that leverage these literals to verify various properties encoded as constraints. Specifically, we introduce two verification queries: if-then rules (ITR) and feature monotonicity (FMO). We benchmark the efficiency of our verification scheme and demonstrate its practical utility in real-world scenarios.