arXiv cs.CL· Sungjun Lim, Heedong Kim, Andrew Lee, Kyungwoo Song·· 12 小时前AI 评分26
Geometry-Adaptive Explainer：分布偏移下基于字典的可解释性保真方法
Geometry-Adaptive Explainer for Faithful Dictionary-Based Interpretability under Distribution Shift
AI 导读
研究提出 Geometry-Adaptive Explainer（GAE），通过将字典与 OOD 活跃子空间重新对齐，缓解分布偏移导致的字典错位问题，且仅需无标注 OOD 激活、无需梯度更新。
正文 · 原文
Abstract:Mechanistic interpretability aims to explain a model's behavior by identifying causally responsible internal structures. Dictionary-based explainers such as sparse autoencoders and transcoders are a primary tool, but their faithfulness under out-of-distribution (OOD) shift has received little systematic attention. We show that distribution shift rotates the subspace that the model actively uses, misaligning the explainer's dictionary trained on in-distribution (ID) activations. We formalize this misalignment as the faithfulness gap, a geometric distance between the ID dictionary and the OOD-active subspace, and show that it bounds excess OOD projection loss. To reduce this gap, we propose the Geometry-Adaptive Explainer (GAE), which realigns the dictionary with the OOD-active subspace while preserving the original sparse feature activations. This requires only unlabeled OOD activations and no gradient updates. For the adapted dictionary, we also bound the excess projection loss caused by any remaining faithfulness gap. Empirically, GAE even matches or surpasses all training-based baselines in causal faithfulness across multiple models and OOD settings.
| Subjects: | Machine Learning (cs.LG); Computation and Language (cs.CL) |
| Cite as: | arXiv:2605.21849 [cs.LG] |
| (or arXiv:2605.21849v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2605.21849 arXiv-issued DOI via DataCite |
Submission history
From: Sungjun Lim [view email]
[v1]
Thu, 21 May 2026 00:46:01 UTC (554 KB)
[v2]
Wed, 4 Nov 2026 05:43:59 UTC (661 KB)
来源：arXiv cs.CL · arxiv.org