Files
myWiki/concepts/prediction-invariant-intervals.md

1.5 KiB
Raw Permalink Blame History

title, created, updated, type, tags, sources
title created updated type tags sources
预测不变区间 (Prediction-Invariant Intervals) 2026-07-04 2026-07-04 concept
certification
interval
prediction
semantics
arXiv:2606.18839

预测不变区间 (Prediction-Invariant Intervals)

语义鲁棒性认证的输出:将语义 extent 范围 [\varphi_a, \varphi_{a'}] 切分为若干子区间,在每个区间内 VLM 预测类别保持不变。

计算过程

  1. 闭式 margin:将语义变换 \gamma(\varphi) 代入 pairwise margin m_{c,c'}(\varphi) = A_{c,c'} \cos\varphi + B_{c,c'} \sin\varphi + C_{c,c'}
  2. 求解穿越点m_{c,c'}(\varphi) = 0 的闭式解
  3. 收集排序:所有类别对的穿越点集合 \{\varphi_\ell\}_{\ell=0}^L
  4. 区间划分(\varphi_\ell, \varphi_{\ell+1}) 内预测恒定

证书输出

\mathcal{S} = \{((\varphi_\ell, \varphi_{\ell+1}), y_\ell) : \ell = 0, \ldots, L-1\}
  • 每个区间标注预测类别 y_\ell
  • 区间边界对应类别翻转点class flip

聚合指标:预测不变概率

\mathbb{P}(f(\gamma(\varphi)) = \hat{y}) = \frac{1}{\varphi_{a'} - \varphi_a} \sum_{\ell: y_\ell = \hat{y}} (\varphi_{\ell+1} - \varphi_\ell)

度量整个语义 extent 范围内预测保持不变的总比例

参考