Metraipolis › archive › dc_11cd26828dfa
Active Budget Can Kill Sensitivity: Diagnosing and Repairing TopK Sparse Autoencoder Reliability — arXiv:2609.37857
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.37857
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T16:24:13.992Z
- SHA-256 of the archived text
d8c7ee2184da3a48988510dea565f5978ef8f4c451e5885b9941ec7f0b35d506- Stable URL of this page
https://metraipolis.com/documento/dc_11cd26828dfa
How to cite
Active Budget Can Kill Sensitivity: Diagnosing and Repairing TopK Sparse Autoencoder Reliability — arXiv:2609.37857 — arXiv (en), https://arxiv.org/abs/2609.37857 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_11cd26828dfa (Metraipolis archive dc_11cd26828dfa, sha256 d8c7ee2184da3a48988510dea565f5978ef8f4c451e5885b9941ec7f0b35d506)
BibTeX
@misc{dc11cd26828dfa,
title = {Active Budget Can Kill Sensitivity: Diagnosing and Repairing TopK Sparse Autoencoder Reliability — arXiv:2609.37857},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.37857},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_11cd26828dfa; sha256 d8c7ee2184da3a48988510dea565f5978ef8f4c451e5885b9941ec7f0b35d506},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.37857. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Sparse autoencoders (SAEs) are increasingly scaled to wider dictionaries to recover fine-grained structure from large language model activations. However, a feature is useful for interpretation only if it remains a stable unit of analysis when the same meaning is expressed in different surface forms. We study this reliability question for TopK SAEs via feature sensitivity. Experiments demonstrate that scaling selectively reduces the sensitivity of rare features, while common features remain comparatively stable. A controlled width\(\times k\) factorial experiment identifies the active budget k as the root cause: the degradation arises from the selection boundary rather than dictionary width alone. We attribute this failure to the geometry of TopK selection. The active margin, the distance to the cutoff, predicts feature loss without thresholds. Guided by this margin diagnosis, we introduce pairwise rank stabilization. Our method targets ordering failures at the cutoff and improves rare-feature sensitivity by \(8.83\) percentage points, while keeping reconstruction and alive-feature coverage near the baseline. Overall, our results suggest that wide TopK SAEs should be evaluated not only by reconstruction, sparsity, and feature count, but also by feature reliability under semantic variation and boundary geometry for stable interpretability. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-30 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.37857 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-07T15:54:53.664Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-30?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).