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 2605.03058cs.LGcs.AI
收藏

基于对比分层消融的大语言模型神经元锚定规则提取

Neuron-Anchored Rule Extraction for Large Language Models via Contrastive Hierarchical Ablation

Università della Svizzera italiana(瑞士意大利大学)

Francesco Sovrano, Gabriele Dominici, Marc Langheinrich

AI总结:

提出MechaRule方法,通过定位稀疏激动剂激活将规则提取锚定在LLM电路中,利用自适应组测试和置信引导剪枝,以极低代价高召回率识别关键神经元,并在算术和越狱任务中验证其有效性。

摘要:

可解释AI的一个核心目标是符号化地表达大语言模型(LLM)的决策逻辑,并将其锚定在内部机制中。现有的规则提取方法通常学习非锚定的符号代理,而机械可解释性将行为与神经元联系起来,但通常需要手工假设和昂贵的干预。我们提出MechaRule,一种通过定位稀疏激动剂激活(其消融会破坏规则相关行为)将规则提取锚定在LLM电路中的流程。MechaRule基于两个发现。首先,在固定的基线/翻转机制下,稀疏激动剂效应可能表现出“超越”:少数高效应的激活在较大组中仍可检测到,主导较弱效应,并翻转许多相同的示例。在这种机制下,使用置信引导的保守剪枝的自适应组测试,当k << N为激动剂时,需要对N个候选进行O(k log(N/k) + k)次干预。其次,在与接近忠实规则行为对齐的数据分割上,激动剂的定位更可靠;谱分割提供了无规则的备选方案,而不忠实的分割会降低定位效果。实验上,在算术和越狱任务中,MechaRule在匹配的暴力验证中召回97.0%的最高效应激动剂,平均仅消耗完全消融成本的2.14%。消融定位的激动剂消除了97.6–100.0%的合格正确算术答案和越狱,并可纠正算术错误或诱导越狱,分别高达72.8%和32.5%。

英文摘要:

A central goal of explainable AI is to express large language model (LLM) decision logic symbolically and ground it in internal mechanisms. Existing rule-extraction methods usually learn ungrounded symbolic surrogates, while mechanistic interpretability links behavior to neurons but often requires hand-crafted hypotheses and costly interventions. We introduce MechaRule, a pipeline that grounds rule extraction in LLM circuits by localizing sparse agonist activations whose ablation disrupts rule-related behavior. MechaRule rests on two findings. First, in a fixed baseline/flip regime, sparse agonist effects can exhibit overtopping: a few high-effect activations remain detectable within larger groups, dominate weaker ones, and flip many of the same examples. In such regimes, adaptive group testing with confidence-guided conservative pruning requires O(k log(N/k) + k) interventions over N candidates when k << N are agonists. Second, agonists are localized more reliably on data splits aligned with close-to-faithful rule behavior; spectral splits provide a rule-free fallback, whereas unfaithful splits degrade localization. Empirically, on arithmetic and jailbreaking, MechaRule recalls 97.0% of highest-effect agonists in matched brute-force validations at only 2.14% of exhaustive-ablation cost on average. Ablating the localized agonists eliminates 97.6--100.0% of eligible correct arithmetic answers and jailbreaks, and can correct arithmetic errors or induce jailbreaks by up to 72.8% and 32.5%.

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。

↑