Reproduce “Diverse Reasoning Paths Matter: Symbolic Equivalence Augmentation for Enhancing RL Exploration” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Xiaoou Liu, Tiejin Chen, Hanyang Chen, Yuan Ling, Hua Wei

EquivAug creates structurally different, correct reasoning paths by translating solutions into symbolic form, applying meaning-preserving rewrites, checking them with solvers, and reconstructing natural-language derivations. In FOLIO, Qwen2.5-3B reached 78.27% exact-match accuracy after GRPO, versus 72.51% with paraphrasing. Across GSM8K and FOLIO, it led post-GRPO accuracy in six tested settings; larger-scale effects remain unknown.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.