Robust Reasoning Under Noisy Supervision: Online Label Refinement Enables LLMs to Self-Correct in Mislabeled Data

Introduction / Main Floor: Robust Reasoning Under Noisy Supervision: Online Label Refinement Enables LLMs to Self-Correct in Mislabeled Data

This paper systematically analyzes the noisy label mechanism in RLVR training, proposes an online label refinement method called OLR, gradually corrects mislabels through majority voting and dynamic consistency detection, and significantly improves model robustness even under noise ratios as high as 90%.