arXiv Artificial Intelligence· Haotian Liu, Yingjian Chen, Xiaohang Xu, Lang Huang, Yusuke Iwasawa, Yutaka Matsuo, Irene Li·· 20 小时前AI 评分24
DDRG：通过推理图对齐跨采样轨迹细粒度纠正 LLM 推理
Beyond Answer Voting: Fine-Grained Correction of LLM Reasoning through Graph Alignment across Sampled Traces
AI 导读
DDRG（Disagreement-Driven Verification Guided by Reasoning Graphs）通过图对齐暴露 LLM 推理轨迹中的声明级分歧，并将局部冲突交给不看答案的验证器，仅在验证器支持的驳斥触及多数答案支持路径时触发一次定向修复，否则返回原始自一致性投票结果。
来源：arXiv Artificial Intelligence · arxiv.org