arXiv:2606.06333cs.LGcs.AI2026-06被引 1

改进稀疏自编码器,让大模型特征更少分裂、更易解释。

Subspace-Aware Sparse Autoencoders for Effective Mechanistic Interpretability

论文配图:Subspace-Aware Sparse Autoencoders for Effective Mechanistic Interpretability
图 1 · 摘自论文原文
  • 用子空间替代单向解码器,支持多维特征表示。
  • 实验显示特征分裂减少一半,且仅需一半训练量。
  • 适合关注模型内部机制解释的研究者。

稀疏自编码器(SAEs)广泛用于大语言模型的机制可解释性研究,但其传统形式为每个潜在特征分配单一解码方向,隐含假设特征为一维。我们证明这一假设与模型特征的真实多维结构不符,会通过两种机制导致特征分裂。几何上,用单方向解码器以误差ε重建内在维度$d_i \ge 2$的特征,所需原子数量呈指数级增长;从端到端优化角度看,这种分裂不仅是可能的,更是被主动偏好。我们证明存在一条从真实$d_i$维基底到更低$\ ext{ℓ}_1$正则化目标风险的连续路径,其下降方向驱动字典进入指数级分裂状态。因此,一个连贯特征被分散至多个近共线的潜在变量中,产生虚假多样性并遮蔽内在几何结构。为此,我们提出子空间感知稀疏自编码器(SASA),以学习的解码子空间替代单向量解码器,通过Top-$s$组门控实现块稀疏,并用核范数正则化自适应每组有效秩。当块大小满足$r \ge d_i$时,单个组不仅能完整表示特征片,还是SASA目标的全局最小值。该整合使样本复杂度由指数降为多项式——在每次训练激活代价高昂的LLM场景下具有决定性优势。实验证明,在GPT-2和Mistral-7B上,SASA显著降低特征分裂与吸收,提升单义性与可解释性,且在约一半的令牌预算下达到或超越标准SAEs性能。

原文摘要 · Abstract (English)

Sparse Autoencoders (SAEs) are widely used for mechanistic interpretability in large language models, yet their formulation assigns each latent feature a single decoder direction, implicitly assuming features to be one-dimensional. We show that this assumption mismatches with the multi-dimensional structure of model features, provably inducing feature splitting through two distinct mechanisms. Geometrically, reconstructing a feature of intrinsic dimension $d_i \ge 2$ to error $\varepsilon$ with single-direction decoders forces a number of atoms that is exponential in $d_i$. From an end-to-end optimization perspective, this splitting is not merely possible but actively preferred. We prove that there exists a continuous path from the true $d_i$-dimensional basis to a strictly lower risk of the $\ell_1$-regularized SAE objective, whose descent directions drive any trained dictionary into that exponential regime. A single coherent feature is therefore fragmented across many near-collinear latents, producing spurious multiplicity and obscuring the intrinsic geometry. Motivated by this, we introduce Subspace-Aware Sparse Autoencoders (SASA), which replace single-vector decoders with learned decoder subspaces, enforce block sparsity via Top-$s$ group gating, and adapt each group's effective rank with a nuclear-norm regularizer. We then show that once the block size satisfies $r \ge d_i$, a single group not only can represent the entire feature slice but is the global minimizer of the SASA objective. This consolidation yields a sample complexity polynomial in $d_i$ rather than exponential -- a decisive advantage given that every training activation costs an LLM forward pass. Empirically, on GPT-2 and Mistral-7B, SASA reduces feature splitting and absorption, improves monosemanticity and interpretability, and matches or exceeds standard SAEs while training on roughly half the token budget.

可解释性稀疏编码大模型

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。