标题： 集成多个在VLLM内部表示上训练的幻觉检测器 显示英文标题

标题： Ensembling Multiple Hallucination Detectors Trained on VLLM Internal Representations

摘要： 本文介绍了我们团队y3h2在KDD Cup 2025的Meta CRAG-MM挑战赛中获得第5名的解决方案。 CRAG-MM基准是一个专注于图像事实性问题的视觉问答（VQA）数据集，包括第一视角图像。 比赛是根据VQA准确性进行竞争的，由基于大语言模型的自动评估器进行评判。 由于错误答案会导致负分，我们的策略集中在减少视觉语言模型内部表示中的幻觉。 具体来说，我们使用隐藏状态和特定注意力头的输出训练了基于逻辑回归的幻觉检测模型。 然后我们采用了这些模型的集成。 结果是，虽然我们的方法牺牲了一些正确答案，但它显著减少了幻觉，并使我们能够在最终排行榜上进入前列。 有关实现细节和代码，请参考https://gitlab.aicrowd.com/htanabe/meta-comprehensive-rag-benchmark-starter-kit。 显示英文摘要

摘要： This paper presents the 5th place solution by our team, y3h2, for the Meta CRAG-MM Challenge at KDD Cup 2025. The CRAG-MM benchmark is a visual question answering (VQA) dataset focused on factual questions about images, including egocentric images. The competition was contested based on VQA accuracy, as judged by an LLM-based automatic evaluator. Since incorrect answers result in negative scores, our strategy focused on reducing hallucinations from the internal representations of the VLM. Specifically, we trained logistic regression-based hallucination detection models using both the hidden_state and the outputs of specific attention heads. We then employed an ensemble of these models. As a result, while our method sacrificed some correct answers, it significantly reduced hallucinations and allowed us to place among the top entries on the final leaderboard. For implementation details and code, please refer to https://gitlab.aicrowd.com/htanabe/meta-comprehensive-rag-benchmark-starter-kit.