跳到正文
arXiv cs.AI· Zane Fu, Yanzhao Zheng, Jinze Liu, Ziting Qin, Qianshan Wei, Jihuai Zhu, Yipeng Yu, Baohua Dong, Hangcheng Zhu, Ruohui Huang·· 4 小时前AI 评分39

研究揭示 On-Policy Distillation 中的视觉后门传播风险

As Student Rollouts Continue, Visual Backdoors Keep Growing

AI 导读

研究挑战了 On-Policy Distillation(OPD)天然抵抗后门传播的直觉,发现视觉后门在 OPD 中低毒化率下几乎不出现,但一旦超过临界阈值,攻击成功率会迅速逼近 100%。后门传播不依赖显式恶意回答,而由教师模型持续累积的异常概率质量驱动;作者据此提出一种零显式触发器注入的触发设计。

正文 · 原文

View PDF HTML (experimental)

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 recently emerged as an important post-training paradigm for Vision-Language Models (VLMs). Unlike traditional offline distillation, OPD does not require teacher-generated responses. Instead, it uses the teacher's probability distributions to supervise response trajectories generated online by the student. Since the initial student model typically generates benign responses and receives no explicit supervision from malicious target responses, OPD appears to offer some inherent resistance to backdoor propagation.
In this paper, we challenge this intuition and reveal a more subtle risk of visual backdoor propagation in OPD. Through experiments across multiple model families and datasets, together with theoretical analysis, we identify three main findings. First, unlike the relatively gradual propagation observed in traditional offline distillation, backdoors rarely emerge in OPD at low poisoning ratios. However, the attack success rate rapidly approaches 100% once a critical threshold is exceeded. Second, backdoor propagation does not depend on explicit malicious responses. Instead, it is driven by the continuous accumulation of anomalous probability mass assigned by the teacher. Finally, based on these findings, we propose a novel trigger design that enables visual backdoor propagation with zero explicit trigger injection. These findings further reveal an overlooked mechanism of backdoor propagation in OPD and highlight its potential security risks.
Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
Cite as: arXiv:2611.22446 [cs.CV]
  (or arXiv:2611.22446v1 [cs.CV] for this version)
  https://doi.org/10.48550/arXiv.2611.22446

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Zane Fu [view email]
[v1] Tue, 17 Nov 2026 09:15:01 UTC (343 KB)

来源:arXiv cs.AI · arxiv.org