揭示Transformer中LRP与留一法差异的根源，提出改进方案。
When LRP Diverges from Leave-One-Out in Transformers
- 发现AttnLRP的双线性传播违反实现不变性
- 通过值矩阵反向传播显著提升与留一法对齐度
- 适合关注模型可解释性与归因准确性的研究者
留一法（LOO）虽直观但计算成本高。层间重要性传播（LRP）虽具潜力，但在现代Transformer中的理论基础尚未充分验证。本文首先证明近期AttnLRP采用的双线性传播规则违反实现不变性公理，并在线性注意力层中实证确认。其次，重新审视CP-LRP作为诊断基线，发现绕过softmax层进行重要性传播——仅通过值矩阵反向传播——能显著提升与LOO的对齐度，尤其在中间到后期变压器层中。总体表明，(i) 双线性因子敏感性和 (ii) softmax传播误差可能共同导致LRP在变压器中难以逼近LOO。
原文摘要 · Abstract (English)
Leave-One-Out (LOO) provides an intuitive measure of feature importance but is computationally prohibitive. While Layer-Wise Relevance Propagation (LRP) offers a potentially efficient alternative, its axiomatic soundness in modern Transformers remains largely under-examined. In this work, we first show that the bilinear propagation rules used in recent advances of AttnLRP violate the implementation invariance axiom. We prove this analytically and confirm it empirically in linear attention layers. Second, we also revisit CP-LRP as a diagnostic baseline and find that bypassing relevance propagation through the softmax layer -- backpropagating relevance only through the value matrices -- significantly improves alignment with LOO, particularly in middle-to-late Transformer layers. Overall, our results suggest that (i) bilinear factorization sensitivity and (ii) softmax propagation error potentially jointly undermine LRP's ability to approximate LOO in Transformers.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。