arXiv：cs.LG· Md Nurul Absar Siddiky, Liuwan Zhu, Yingfei Dong·· 3 小时前AI 评分47
频率不等于敏感度：在稀疏 MoE LLM 中识别安全敏感专家
Frequency Is Not Sensitivity Identifying Safety-Sensitive Experts in Sparse MoE LLM
AI 导读
研究提出用 router-gradient 敏感度替代激活频率来筛选稀疏 MoE 大语言模型中需抑制的专家，在五种 MoE 架构、500 条良性加 500 条恶意提示上排序，并在 100 条留出恶意提示上测试拒绝率。
正文
Abstract:Suppressing a small set of routed experts can weaken the safety behavior of a sparse Mixture-of-Experts (MoE) language model without retraining. Which experts to suppress is therefore a security question, and the usual answer is activation frequency, but frequency measures use, not influence. We test an alternative: router-gradient sensitivity, the sensitivity of the sequence loss to the gate weights that select an expert. Across five MoE architectures, we rank experts by each signal on 500 benign and 500 malicious prompts and measure refusal on 100 held-out malicious prompts under two budgets: equal expert counts and equal nominal malicious routing traffic (1%-5%). Under each of the two budgets, router-gradient selection reduces refusals more than activation in 24 of 25 conditions, and more than a ten-trial random mean in all 25. The largest effect is in OLMoE, where refusals fall from 34 to 9 of 100 prompts (73.53% relative) with no degraded outputs, indicating substantive compliance rather than broken generation. After matching expert counts in every layer, gradient selection still produces greater refusal reduction than activation in 23 of 25 conditions, with two ties. An exploratory cross-model analysis links larger malicious-versus-benign concentration gaps to greater peak gradient effects (rho = 0.90; exact two-sided p = 0.083, n = 5). Together, the results support gradient selection under the tested budgets.
| Subjects: | Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.02910 [cs.AI] |
| (or arXiv:2610.02910v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.02910 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Md Nurul Absar Siddiky [view email]
[v1]
Fri, 2 Oct 2026 07:00:49 UTC (239 KB)
来源：arXiv：cs.LG · arxiv.org