arXiv cs.AI· Masha Fedzechkina, Eleonora Gualdoni, Rita Ramos, Sinead Williamson·· 10 天前AI 评分28
模型 top-k logits 会泄露图像查询中的任务无关信息
What do your logits know?
AI 导读
研究以视觉语言模型为测试平台，首次系统比较了残差流经两个自然瓶颈压缩后各表征层保留的信息：tuned lens 得到的低维投影，以及最可能影响模型答案的最终 top-k logits。结果显示，即便是模型 top logit 值定义的易获取瓶颈，也会泄露图像查询中与任务无关的信息，某些情况下泄露量与完整残差流的直接投影相当。
正文 · 原文
Abstract:Recent work has shown that probing model internals can reveal a wealth of information not apparent from the model generations. This poses a risk of unintentional or malicious information leakage, where model users are able to learn information that the model owner assumed was inaccessible. Using vision-language models as a testbed, we present the first systematic comparison of information retained at different representational levels as it is compressed from the rich information encoded in the residual stream through two natural bottlenecks: low-dimensional projections of the residual stream obtained using tuned lens, and the final top-k logits most likely to impact model's answer. We show that even easily accessible bottlenecks defined by the model's top logit values can leak task-irrelevant information present in an image-based query, in some cases revealing as much information as direct projections of the full residual stream.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2604.09885 [cs.AI] |
| (or arXiv:2604.09885v3 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2604.09885 arXiv-issued DOI via DataCite |
Submission history
From: Masha Fedzechkina [view email]
[v1]
Fri, 10 Apr 2026 20:24:38 UTC (20,784 KB)
[v2]
Fri, 2 Oct 2026 00:50:16 UTC (21,790 KB)
[v3]
Mon, 5 Oct 2026 20:35:10 UTC (21,792 KB)
来源：arXiv cs.AI · arxiv.org