标签
提出 Moose,一种神经符号方法,将 OWL 2 EL 本体编译为句子决策图(SDD),用于可微加权模型计数,从而在部分监督下实现潜概念学习,并首次在 OWL EL 设置中提供推理捷径分析。
This paper analyzes reasoning shortcuts in neurosymbolic systems through symmetry and automorphism groups, reveals flaws in a prior framework, proves complexity-theoretic results, and empirically validates the corrected theory on benchmarks like CLE4EVR and rsbench.