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 2602.12665cs.AI
收藏

评估参数化逻辑问题上推理模型的鲁棒性

Evaluating Robustness of Reasoning Models on Parameterized Logical Problems

CRIL UMR 8188, Univ Artois, CNRS, France(CRIL UMR 8188,阿维尼翁大学,法国国家科学研究中心)

Naïm Es-sebbani, Esteban Marquer, Yakoub Salhi, Zied Bouraoui

AI总结:

本文提出了一种针对2-SAT的诊断基准,用于评估推理模型在参数化逻辑问题上的鲁棒性,通过隔离不同能力与失败模式,揭示模型在结构干预下的性能变化。

摘要:

逻辑为评估基于大语言模型的推理器提供了受控的测试环境,但标准的SAT类基准往往将表面难度(长度、措辞、子句顺序)与实际上决定可满足性的结构现象混淆。我们引入了一个针对2-SAT的诊断基准,该基准基于参数化家族的结构2-CNF公式构建,其中可满足性由蕴含图来表征,可以沿可解释的轴进行调节。我们的生成器隔离了不同的能力和失败模式:(i) 存在矛盾循环的不可满足核心,具有可控的大小和不平衡性;(ii) 具有规定比例自由变量的可满足实例,以控制解的多重性;(iii) 手植的骨架,调节传播;(iv) 晚期桥接子句,将否则单调的区域联系起来,以探测对顺序和修订的敏感性;(v) 对称/复制变体,测试在重命名和冗余结构下的抽象能力。我们评估基于LLM的推理器在决策准确性和赋值有效性上的表现,并量化在语义保持扰动下(如子句重排、填充子句和变量重命名)的鲁棒性。在各模型中,我们观察到在目标结构干预下,即使表面统计保持固定,性能也出现明显的转折,揭示了在总体SAT准确性不可见的脆弱性区域。

英文摘要:

Logic provides a controlled testbed for evaluating LLM-based reasoners, yet standard SAT-style benchmarks often conflate surface difficulty (length, wording, clause order) with the structural phenomena that actually determine satisfiability. We introduce a diagnostic benchmark for 2-SAT built from parameterized families of structured 2--CNF formulas, where satisfiability is characterized by the implication graph and can be tuned along interpretable axes. Our generators isolate distinct competencies and failure modes: (i) contradiction-cycle UNSAT cores with controllable size and imbalance, (ii) SAT instances with a prescribed fraction of free variables to control solution multiplicity, (iii) planted backbones that modulate propagation, (iv) late bridge clauses that couple otherwise monotone regions to probe sensitivity to ordering and revision, and (v) symmetry/duplication variants that test abstraction under renaming and redundant structure. We evaluate LLM-based reasoners on decision accuracy and assignment validity, and quantify robustness under semantics-preserving perturbations such as clause reordering, filler clauses, and variable renaming. Across models, we observe sharp performance transitions under targeted structural interventions even when surface statistics are held fixed, revealing brittleness regimes that are invisible to aggregate SAT accuracy.

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。

↑