Neural Decision-Propagation for Efficient Answer Set Programming
A new method called decision-propagation (DProp) computes stable models by alternating falsity decisions and truth propagations, capturing stable model semantics. Its differentiable extension, Neural DProp (NDProp), integrates neural computation for decisions and fuzzy evaluation for propagations, enabling learning of decision heuristics and neuro-symbolic integration. Experiments show NDProp efficiently learns to compute stable models, improving scalability over classical solvers. The work addresses a bottleneck in neuro-symbolic AI by replacing traditional ASP solvers with a neural approach.
Key facts
- DProp alternates falsity decisions and truth propagations.
- DProp computations capture stable model semantics.
- NDProp is a differentiable extension of DProp.
- NDProp uses neural computation for decisions and fuzzy evaluation for propagations.
- NDProp is evaluated for learning decision heuristics and neuro-symbolic integration.
- Results show NDProp can learn to efficiently compute stable models.
- The method improves scalability over classical ASP solvers.
- The work is published on arXiv with ID 2605.01797.
Entities
Institutions
- arXiv