GUT: Quantifying and Optimizing the Reasoning Uncertainty of LLMs via Graph Complexity
2026-09-07 12:00Science🔥 40.2 heat score
1sources
1days unfolding
40.2heat score
1mentions
SummaryAI generated
On September 7, 2026, arXiv cs.AI published research results titled GUT. This study proposed a new method to quantify and optimize the inference uncertainty of large language models (LLMs) through graph complexity.
近日，研究团队提出基于图复杂度的不确定性（GUT）方法以量化并优化大语言模型（LLMs）的推理不确定性。该方法将每条推理链的潜在分支表征为有向无环图，构建包含量化模块（GUT-Q）与优化模块（GUT-O）的双模块体系。其中，GUT-Q 通过近似推理空间复杂度来衡量不确定性，GUT-O 则将负不确定性作为强化学习中的奖励函数以实施优化。实验在四种大语言模型和五个数据集上验证了 GUT 的有效性。