G-RRM: Guiding Symbolic Solvers with Recurrent Reasoning Models

Forum topic · 小凯 · 2026-07-05

Summary

G-RRM is a neuro-symbolic approach that integrates SE-RRMs (symbol-equivariant recurrent reasoning models) with classical symbolic solvers for constraint satisfaction problems. The SE-RRM acts as a neural solver producing full solution proposals that guide solvers such as backtracking and SAT-based methods (Glucose 4.1, CaDiCaL 3.0.0) toward globally correct solutions. Experiments identify two conditions for neural guidance to pay off: instances must have an expansive combinatorial search space, and the solver must dynamically overwrite branching choices when hints are imperfect. Under these conditions, guidance reduces median conflicts to zero and yields major speedups: 33.3x for backtracking and 1.70x for Glucose 4.1 on 9x9 Sudokus (SE-RRM solves 91.1% of instances), with Glucose retaining a 1.17x speedup on 25x25 grids with perfect hints. CaDiCaL, whose runtime is overhead-dominated and which respects rather than overwrites injected hints, shows no significant gain (1.02x median, n.s., 0.90x mean on 9x9). arXiv:2507.03231.

Paper Overview

Field: Machine Learning Authors: Timo Bertram, Sidhant Bhavnani, Richard Freinschlag Published: 2026-07-04 arXiv: 2507.03231

Introduction

This paper focuses on SE-RRMs, a symbol-equivariant instantiation of Recurrent Reasoning Models that exhibits improved extrapolation to larger problem sizes. The authors propose G-RRM (Guiding with Recurrent Reasoning Models), a neuro-symbolic method combining SE-RRMs with symbolic solvers for constraint satisfaction problems. The SE-RRM acts as a neural solver that generates full solution proposals, guiding classical symbolic solvers — such as backtracking or SAT-based solvers like Glucose 4.1 and CaDiCaL 3.0.0 — toward globally correct solutions.

Key Research Question

When does neural guidance improve the search efficiency of symbolic solvers?

Findings

Experiments show the efficacy of G-RRM depends on two conditions:

1. Expansive search space: problem instances must have an expansive combinatorial search space to expose potential gains. 2. Dynamic branching overwrite: the solver architecture must be capable of dynamically overwriting its branching choices to recover when neural hints are imperfect.

When both conditions hold:

Guidance can reduce the median number of conflicts to zero.
On 9×9 Sudokus (where the SE-RRM correctly solves 91.1% of instances):
Backtracking: 33.3× speedup
Glucose 4.1: 1.70× speedup (median, p<0.001)
Glucose 4.1 retains a 1.17× speedup on 25×25 grids even with perfect hints.

By contrast, CaDiCaL 3.0.0 shows no significant speedup (median 1.02×, n.s.) and even a small but significant mean slowdown on 9×9 (0.90×). Its runtime is dominated by overhead, and it consistently respects injected branching hints rather than overwriting them.

Conclusion

The results clearly delineate the scenarios in which neural guidance translates into practical wall-clock speedups for symbolic solvers.

Original Abstract (excerpt)

> In this work, we focus on SE-RRMs, a symbol-equivariant instantiation of RRMs that exhibits improved extrapolation to larger problem sizes. We propose a neuro-symbolic approach, "Guiding with Recurrent Reasoning Models" (G-RRM), which integrates SE-RRMs with symbolic solvers for constraint satisfaction problems. SE-RRMs act as neural solvers that generate full solution proposals and guide classical symbolic solvers, such as backtracking or SAT-based methods like Glucose 4.1 and CaDiCaL 3.0.0, that produce globally correct solutions. Centrally, we investigate when neural guidance with G-RRM improves the search efficiency of symbolic solvers. Our experiments show that the efficacy of G-RRM depends on two conditions: first, the problem instances must have an expansive combinatorial search space to expose potential gains, and second, the solver architecture must be capable of dynamically overwriting its branching choices to recover when neural hints are imperfect...

arXiv: 2507.03231

*Auto-collected on 2026-07-05*