余弦相似度不是证据：量化下可解释性迁移的噪声底测量
原标题：Cosine Similarity Is Not Evidence: Measuring the Noise Floor of Interpretability Transfer Under Quantization
AI 摘要
该论文指出，在量化模型的可解释性迁移研究中，仅报告余弦相似度等尺度不变统计量而不提供其噪声底（noise floor）是不足以作为证据的。作者针对差值均值方向估计器推导出分割半信度噪声底的闭式解，该噪声底仅由一个无量纲数决定，并测量了真实激活上的类别分离度。在Qwen2.5-1.5B-Instruct上，INT4量化下拒绝方向发生旋转且偏移超过估计器自身噪声，而INT8下未检测到移动，但这并不构成等价性声明。作者还指出尺度不变统计量无法区分迁移决策变量的平移与衰减，并给出了仅需一次前向传播的报告建议。
正文节选
Cosine Similarity Is Not Evidence: Measuring the Noise Floor of Interpretability Transfer Under Quantization Abstract A statistic reported without the quantity needed to interpret it is not evidence. We develop that thesis for a concrete practice in AI safety. Interpretability artifacts are calibrated on full-precision weights, deployed on quantized ones, and certified as surviving the change by scale-invariant statistics (cosine similarity, correlation, AUROC) that are reported without their no