用谎言教学：基于合成负样本的课程 DPO 用于幻觉检测
Teaching with Lies: Curriculum DPO on Synthetic Negatives for Hallucination Detection
AI总结：
针对大模型幻觉检测难题，提出用合成幻觉作负样本、结合课程学习的DPO方法，训练HaluCheck模型，在困难基准上提升高达24%，零样本下超越更大模型。
摘要：
将大型语言模型（LLMs）对齐以准确检测幻觉仍然是一项重大挑战，因为幻觉文本具有复杂的性质。认识到幻觉样本通常比传统负样本表现出更高的欺骗性质量，我们在 DPO 对齐过程中使用这些精心设计的幻觉作为负样本。我们的方法融入了课程学习策略，逐步将训练从基于独立事实检查模型概率分数下降最大的简单样本过渡到逐渐困难的样本。这种结构化的难度缩放确保了稳定和渐进的学习。实验评估表明，我们使用课程 DPO 方法和高质量负样本训练的 HaluCheck 模型，在各种指标上显著提高了模型性能，在 MedHallu 和 HaluEval 等困难基准上实现了高达 24% 的提升。此外，HaluCheck 模型在零样本设置中表现出鲁棒性，在各种基准上显著优于更大的最先进模型。
英文摘要：
Aligning large language models (LLMs) to accurately detect hallucinations remains a significant challenge due to the sophisticated nature of hallucinated text. Recognizing that hallucinated samples typically exhibit higher deceptive quality than traditional negative samples, we use these carefully engineered hallucinations as negative examples in the DPO alignment procedure. Our method incorporates a curriculum learning strategy, gradually transitioning the training from easier samples, identified based on the greatest reduction in probability scores from independent fact checking models, to progressively harder ones. This structured difficulty scaling ensures stable and incremental learning. Experimental evaluation demonstrates that our HaluCheck models, trained with curriculum DPO approach and high quality negative samples, significantly improves model performance across various metrics, achieving improvements of upto 24% on difficult benchmarks like MedHallu and HaluEval. Additionally, HaluCheck models demonstrate robustness in zero-shot settings, significantly outperforming larger state-of-the-art models across various benchmarks.
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。
