Tag
This paper proposes Generative Verification (GenV), a method using generative reward models to detect reference-equivalence failures in autoformalization, addressing vulnerabilities in neurosymbolic systems and improving verification accuracy.
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.