When Right Meets Wrong: Bilateral Context Conditioning with Reward-Confidence Correction for GRPO
This work addresses a bottleneck in training reasoning models for tasks like mathematical reasoning, offering an incremental enhancement to existing GRPO methods.
The paper tackled the problem of Group Relative Policy Optimization (GRPO) overlooking contrastive signals between correct and incorrect reasoning traces, proposing Bilateral Context Conditioning (BICC) and Reward-Confidence Correction (RCC) to leverage this information, resulting in consistent improvements on mathematical reasoning benchmarks.
Group Relative Policy Optimization (GRPO) has emerged as an effective method for training reasoning models. While it computes advantages based on group mean, GRPO treats each output as an independent sample during the optimization and overlooks a vital structural signal: the natural contrast between correct and incorrect solutions within the same group, thus ignoring the rich, comparative data that could be leveraged by explicitly pitting successful reasoning traces against failed ones. To capitalize on this, we present a contrastive reformulation of GRPO, showing that the GRPO objective implicitly maximizes the margin between the policy ratios of correct and incorrect samples. Building on this insight, we propose Bilateral Context Conditioning (BICC), a mechanism that allows the model to cross-reference successful and failed reasoning traces during the optimization, enabling a direct information flow across samples. We further introduce Reward-Confidence Correction (RCC) to stabilize training by dynamically adjusts the advantage baseline in GRPO using reward-confidence covariance derived from the first-order approximation of the variance-minimizing estimator. Both mechanisms require no additional sampling or auxiliary models and can be adapted to all GRPO variants. Experiments on mathematical reasoning benchmarks demonstrate consistent improvements across comprehensive models and algorithms. Code is available at \href{https://github.com/Skylanding/BiCC}{https://github.com/Skylanding/BiCC}.