线性可解释性有效，源于Transformer架构的线性接口约束。
Why Linear Interpretability Works: Invariant Subspaces as a Result of Architectural Constraints
- 通过注意力机制的线性接口，语义特征自然落在不变子空间中。
- 实验验证八项分类任务中类别令牌与语义实例高度对齐。
- 无需标注数据即可零样本识别语义结构，适合模型解释研究。
线性探测器和稀疏自编码器在Transformer表征中持续发现有意义的结构——但为何简单方法能在深层非线性系统中奏效？我们证明这并非偶然现象，而是架构必然结果：Transformer通过线性接口（注意力OV电路、反嵌入矩阵）传递信息，任何经此类接口解码的语义特征必位于上下文不变的线性子空间中。我们形式化提出‘不变子空间必要性’定理，并推导出‘自指性质’：标记符直接提供其关联特征的几何方向，实现零样本语义结构识别，无需标签或训练探测器。在八项分类任务和四个模型族中的实证验证表明，类别标记符与语义相关实例高度对齐。本框架为线性可解释性提供了基于架构的原理性解释，统一了线性探测器与稀疏自编码器。
原文摘要 · Abstract (English)
Linear probes and sparse autoencoders consistently recover meaningful structure from transformer representations -- yet why should such simple methods succeed in deep, nonlinear systems? We show this is not merely an empirical regularity but a consequence of architectural necessity: transformers communicate information through linear interfaces (attention OV circuits, unembedding matrices), and any semantic feature decoded through such an interface must occupy a context-invariant linear subspace. We formalize this as the \emph{Invariant Subspace Necessity} theorem and derive the \emph{Self-Reference Property}: tokens directly provide the geometric direction for their associated features, enabling zero-shot identification of semantic structure without labeled data or learned probes. Empirical validation in eight classification tasks and four model families confirms the alignment between class tokens and semantically related instances. Our framework provides \textbf{a principled architectural explanation} for why linear interpretability methods work, unifying linear probes and sparse autoencoders.
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。