PulseAugur
中
实时 07:57:18
English(EN) As Student Rollouts Continue, Visual Backdoors Keep Growing

视觉后门通过On-Policy Distillation在视觉语言模型中传播

一篇新发布的arXiv论文详细介绍了视觉后门如何在视觉语言模型(VLMs)的On-Policy Distillation(OPD)中传播。与最初的假设相反,OPD本身并不能抵抗后门传播,因为一旦达到关键的投毒阈值,后门就会迅速出现。传播是由教师模型异常概率质量的累积驱动的,而不是明确的恶意响应。研究人员开发了一种新颖的触发器设计,可以在不需要显式触发器注入的情况下促进视觉后门传播,突显了OPD中潜在的安全风险。 AI

影响 强调了视觉语言模型训练中一个微妙的安全风险,该风险可能被利用。

排序理由 学术论文,详细介绍了AI模型训练中的一个安全漏洞。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

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

视觉后门通过On-Policy Distillation在视觉语言模型中传播

本文如何被排名

Signal score
26 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
学术论文,详细介绍了AI模型训练中的一个安全漏洞。[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
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

  1. arXiv cs.AI TIER_1 English(EN) · Zane Fu, Yanzhao Zheng, Jinze Liu, Ziting Qin, Qianshan Wei, Jihuai Zhu, Yipeng Yu, Baohua Dong, Hangcheng Zhu, Ruohui Huang ·

    随着学生推广的继续,视觉后门持续增长

    arXiv:2611.22446v1 Announce Type: cross Abstract: Conventional knowledge distillation can facilitate the propagation of backdoors from teacher models to student models. However, how this risk manifests in On-Policy Distillation (OPD) remains insufficiently studied. OPD has recent…