Graph Neural Reasoning May Fail in Certifying Boolean Unsatisfiability
This work addresses a critical bottleneck in applying GNNs to complex predicate logic reasoning, highlighting potential failures that could impact AI systems relying on logical inference.
The paper investigates the limitations of graph neural networks (GNNs) in certifying Boolean unsatisfiability (UNSAT), suggesting that GNNs may fail in learning logical reasoning tasks that involve UNSAT as a sub-problem, based on conjectures and evidence presented.
It is feasible and practically-valuable to bridge the characteristics between graph neural networks (GNNs) and logical reasoning. Despite considerable efforts and successes witnessed to solve Boolean satisfiability (SAT), it remains a mystery of GNN-based solvers for more complex predicate logic formulae. In this work, we conjectures with some evidences, that generally-defined GNNs present several limitations to certify the unsatisfiability (UNSAT) in Boolean formulae. It implies that GNNs may probably fail in learning the logical reasoning tasks if they contain proving UNSAT as the sub-problem included by most predicate logic formulae.