NeuronTune：用于大语言模型中安全-效用平衡对齐的细粒度神经元调制
NeuronTune: Fine-Grained Neuron Modulation for Balanced Safety-Utility Alignment in LLMs
AI总结：
NeuronTune是一种细粒度神经元调制框架，通过定位安全关键与效用保留神经元并动态控制干预范围，实现了大语言模型安全与效用的平衡优化，性能优于现有最先进技术。
摘要：
在大语言模型（LLM）的可靠部署中，在保留效用的同时确保稳健的安全对齐至关重要。然而，现有技术存在根本缺陷：对恶意攻击的稳健性不足、频繁拒绝良性查询、生成文本质量和通用任务性能下降，前两者体现了稳健安全的张力，后者则构成效用损害。我们将这些限制归因于现有方法中粗粒度的层间干预。为解决该问题，我们提出NeuronTune，这是一种细粒度框架，可定位并调制稀疏神经元以实现安全-效用的同时优化。我们的方法首先通过感知攻击的归因在所有层中定位安全关键和效用保留神经元，然后采用元学习自适应调制其激活。关键在于，NeuronTune的干预范围通过神经元包含阈值动态控制，提供了优先安全关键或效用优先需求的灵活机制。大量实验结果表明，我们的方法优于现有最先进技术，在实现卓越模型安全的同时保持了出色的效用。
英文摘要：
Ensuring robust safety alignment while preserving utility is critical for the reliable deployment of Large Language Models (LLMs). However, current techniques fundamentally suffer from intertwined deficiencies: insufficient robustness against malicious attacks, frequent refusal of benign queries, degradation in generated text quality and general task performance, the former two reflecting tensions in robust safety and the latter constituting utility impairment. We attribute these limitations to the coarse-grained layer-wise interventions in existing methods. To resolve this, we propose NeuronTune, a fine-grained framework that pinpoints and modulates sparse neurons to achieve simultaneous safety-utility optimization. Our approach first pinpoints safety-critical and utility-preserving neurons across all layers via attack-aware attribution, then adapts meta-learning to adaptively modulate their activations. Crucially, the intervention scope of NeuronTune is dynamically controlled via neuron inclusion thresholds, providing a flexible mechanism to prioritize either security-critical or utility-priority requirements. Extensive experimental results demonstrate that our method outperforms existing state-of-the-art technologies, achieving superior model safety while maintaining excellent utility.
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。
