arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2601.14044cs.CV
收藏

Weather-R1: 逻辑一致的强化微调用于气象中的多模态推理

Weather-R1: Logically Consistent Reinforcement Fine-Tuning for Multimodal Reasoning in Meteorology

Sun Yat-sen University(中山大学);Guangdong Meteorological Observatory(广东省气象局)

Kaiyu Wu, Pucheng Han, Hualong Zhang, Naigeng Wu, Keze Wang

AI总结:

本文提出Weather-R1,首个具备逻辑忠实性的气象多模态推理VLM,通过逻辑一致性奖励解决强化微调中的自相矛盾推理问题,并在WeatherQA基准上提升9.8个百分点。

摘要:

尽管视觉语言模型(VLMs)展现了日益增强的推理能力,但其在气象领域的应用受到领域差距和推理忠实性差距的限制。具体而言,主流的强化微调(RFT)可能导致自相矛盾推理(Self-Contra),即模型的推理与最终答案矛盾,这在高风险领域不可接受。为解决这些挑战,我们构建了WeatherQA,一个新颖的气象多模态推理基准。我们还提出了逻辑一致的强化微调(LoCo-RFT),通过引入逻辑一致性奖励来解决Self-Contra问题。此外,我们引入Weather-R1,据我们所知,这是首个在气象领域具备逻辑忠实性的推理VLM。实验表明,Weather-R1在WeatherQA基准上比基线提升了9.8个百分点,优于监督微调和RFT,甚至超过了原始Qwen2.5-VL-32B。这些结果突显了我们LoCo-RFT的有效性以及Weather-R1的优越性。我们的基准和代码可在https://github.com/Marcowky/Weather-R1获取。

英文摘要:

While Vision Language Models (VLMs) show advancing reasoning capabilities, their application in meteorology is constrained by a domain gap and a reasoning faithfulness gap. Specifically, mainstream Reinforcement Fine-Tuning (RFT) can induce Self-Contradictory Reasoning (Self-Contra), where the model's reasoning contradicts its final answer, which is unacceptable in such a high-stakes domain. To address these challenges, we construct WeatherQA, a novel multimodal reasoning benchmark in meteorology. We also propose Logically Consistent Reinforcement Fine-Tuning (LoCo-RFT), which resolves Self-Contra by introducing a logical consistency reward. Furthermore, we introduce Weather-R1, the first reasoning VLM with logical faithfulness in meteorology, to the best of our knowledge. Experiments demonstrate that Weather-R1 improves performance on WeatherQA by 9.8 percentage points over the baseline, outperforming Supervised Fine-Tuning and RFT, and even surpassing the original Qwen2.5-VL-32B. These results highlight the effectiveness of our LoCo-RFT and the superiority of Weather-R1. Our benchmark and code are available at https://github.com/Marcowky/Weather-R1.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑