arXiv cs.CL· Haotian Ma, Ziye Mei, Frederic Sala·· 2 天前AI 评分24
稀疏自编码器表征的行为组合性研究：SAE Additivity Residuals 方法
Behavioral Compositionality of Sparse Autoencoder Representations
AI 导读
研究者提出 SAE Additivity Residuals（SAR），用于衡量稀疏自编码器特征在选定交互阶数之上未被解释的行为变化。在 18 组模型与行为设置中，学习到的 SAE 坐标一阶残差均低于保持和不变的随机旋转；但坐标并非完全可加，成对项在所有设置中均改善留出联盟排序。
正文 · 原文
Abstract:Sparse autoencoders (SAEs) represent model activations with sparse, linearly decoded features, but the downstream effects of those features need not compose additively. We introduce SAE Additivity Residuals (SAR), which measure the behavioral variation left unexplained above a chosen interaction order. Under independent feature subset sampling, population SAR is exactly the normalized energy of interactions above that order and bounds coalition ordering error for the optimal low order approximation. To distinguish compositionality associated with the SAE representation from that induced by downstream computation, we compare learned SAE coordinates with sum preserving random rotations of the same selected subspace and find lower first order residuals for the learned coordinates across all 18 model and behavior settings. The coordinates are nevertheless not fully additive: pairwise terms improve held out coalition ordering in every setting and outperform parameter matched models using randomly selected third order terms. In a combined steering experiment, pairwise modeling raises rank fidelity from 0.5874 to 0.7790 and reduces ordering error from 0.1444 to 0.0606. Together, these results show that SAE coordinates support comparatively compositional behavioral descriptions while faithful joint explanations and interventions still require interactions beyond individual feature effects. Code is available at this https URL.
| Comments: | 29 pages, 9 figures. Code: this https URL |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.21261 [cs.LG] |
| (or arXiv:2610.21261v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.21261 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Ziye Mei [view email]
[v1]
Thu, 15 Oct 2026 04:58:23 UTC (165 KB)
来源：arXiv cs.CL · arxiv.org