PulseAugur
中
实时 12:40:21
English(EN) How Fragile Is On-Device Language Model Safety? Localizing Safety-Critical Parameters for Sparse Fault Analysis

设备端 LLM 安全分析揭示稀疏关键参数

研究人员调查了设备端语言模型(特别是 Llama-2-7B-Chat)的安全性,以确定安全关键参数是否集中在稀疏子集中。他们的分析显示,安全敏感性在模型中分布不均,MLP down_proj 持续显示出高敏感性。通过修改 down_proj 层中一小部分权重(0.19%),他们在保持 tinyBenchmarks 基线准确性的同时,实现了对抗性成功率(ASR)的显著提高,这表明了一种针对设备端模型进行分析和保护的靶向方法。 AI

影响 提出了针对设备端 LLM 进行靶向分析和保护的方法,有可能提高边缘 AI 应用的安全性。

排序理由 该集群包含一篇详细介绍 LLM 安全研究结果的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CL 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →

设备端 LLM 安全分析揭示稀疏关键参数

本文如何被排名

Signal score
8 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
该集群包含一篇详细介绍 LLM 安全研究结果的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
paper, safety
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Same-day
Cluster formed today. Ranking reflects the current source set at time of score.

完整方法见我们的编辑标准。

报道来源 [1]

  1. arXiv cs.CL TIER_1 English(EN) · Muhammad Zeeshan Karamat, Christiana Chamon Garcia ·

    设备端语言模型安全有多脆弱?为稀疏故障分析本地化安全关键参数

    arXiv:2610.09000v1 Announce Type: cross Abstract: As small language models (SLMs) are increasingly deployed on resource-constrained and on-device platforms, including as components of agentic systems, the integrity of locally stored model parameters becomes an important safety co…