Reproduce “CoCA: Criterion-Aware Credit Assignment for Rubric-Based Reinforcement Learning” 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
Guangcheng Zhu, Yingfan Ma, Xing Zheng, Haobo Wang, Shenzhi Yang, Kai Tang, Zhengqing Zang, Bowen Song and 2 more
CoCA separates criterion-level credit determination from token-level credit allocation in rubric-based reinforcement learning. Observed outcomes assign signed credit to each criterion, while Jensen–Shannon divergence between criterion-conditioned and ordinary next-token predictions guides where that credit is applied. Across seven medical and science benchmarks, CoCA averages 64.5%, versus 60.3% for standard rubric-based reinforcement learning; these benchmark results do not establish clinical safety.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.