Title: Neural Decision-Propagation for Answer Set Programming

Abstract: The fusion of Answer Set Programming (ASP) with neural networks represents a significant advancement in Neuro-symbolic AI. Although current methodologies enhance ASP's applicability to real-world scenarios, they are constrained by scalability issues due to their reliance on classical solvers within the reasoning pipeline. To address this limitation, we introduce decision-propagation (DProp), a novel technique for calculating stable models that operates by alternating between falsity decisions and truth propagations. Our analysis demonstrates that successful DProp computations effectively embody stable model semantics. Building on this, we present Neural DProp (NDProp), a differentiable variant of DProp that employs neural networks for decision-making and fuzzy logic for propagation evaluations. We assess NDProp’s performance in learning decision heuristics and facilitating neuro-symbolic integration, benchmarking it against existing neuro-symbolic methods. Our findings indicate that NDProp can efficiently learn to compute stable models, thereby enhancing both accuracy and scalability on neuro-symbolic benchmarks.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC