NoisyGRPO: Incentivizing Multimodal CoT Reasoning via Noise Injection and Bayesian Estimation
NoisyGRPO:通过噪声注入和贝叶斯估计激励多模态Co T推理
机构 * ShanghaiTech University(上海科技大学) ; Shanghai Engineering Research Center of Intelligent Vision and Imaging(上海智能视觉与成像工程研究中心) ; Lingang Laboratory(临港实验室)
专题命中 其他推理 :reasoning(title,abstract);CoT(title,abstract);chain-of-thought(abstract)
AI总结 NoisyGRPO通过引入可控噪声增强探索并利用贝叶斯框架建模优势估计,提升多模态大语言模型的泛化能力和鲁棒性,尤其在小规模模型上表现突出。
Comments Accepted by Neurips 2025, Project page is available at https://artanic30.github.io/project_pages/NoisyGRPO/
Journal ref Advances in Neural Information Processing Systems 38 (2026) 124239-124267