English(EN) CORA: Analyzing and bridging thinking-answer gap in Multimodal RLVR via Consistency-Oriented Reasoning Alignment

新的CORA方法弥合了多模态AI中的思维-答案差距

作者 PulseAugur 编辑部 · [2 个来源] · 2026-06-12 17:54

研究人员推出了一种新方法CORA，用于解决多模态大型视觉语言模型（LVLMs）中存在的思维-答案不一致问题。这种不一致性，即推理过程在语义上与最终答案不匹配，在训练和推理过程中一直存在。CORA利用一致性奖励模型和混合奖励优势分解来提高任务性能并确保更忠实的推理过程。 AI

影响通过提高推理过程的忠实度，解决了多模态AI的一个关键挑战，有望带来更可靠的AI输出。

排序理由该集群包含一篇详细介绍多模态AI新方法的学术论文。

AI 生成摘要 · Google Gemini · 来自 2 个来源。我们如何撰写摘要 →

报道来源 [2]

arXiv cs.CL TIER_1 English(EN) · Jiayue Cao, Zhicong Lu, Xuehan Sun, Wei Jia, Hongling Zheng, Changyuan Tian, Zichuan Lin, Wenqian Lv, Nayu Liu · 2026-06-15 04:00

CORA: Analyzing and bridging thinking-answer gap in Multimodal RLVR via Consistency-Oriented Reasoning Alignment

arXiv:2606.14691v1 Announce Type: new Abstract: Reinforcement learning with verifiable rewards (RLVR) has successfully elicited the reasoning capabilities of large language models, motivating its extension to multimodal scenarios. Existing methods primarily focus on improving the…
arXiv cs.CL TIER_1 English(EN) · Nayu Liu · 2026-06-12 17:54

CORA: Analyzing and bridging thinking-answer gap in Multimodal RLVR via Consistency-Oriented Reasoning Alignment

Reinforcement learning with verifiable rewards (RLVR) has successfully elicited the reasoning capabilities of large language models, motivating its extension to multimodal scenarios. Existing methods primarily focus on improving the visual coverage of reasoning traces and mitigat…