arXiv cs.CL· Karl Li·· 11 小时前AI 评分31
稀疏自编码器归因图由 margin 决定
Sparse Autoencoder Attribution Graphs Are Decided at the Margin
AI 导读
研究发现 TopK SAE 的归因图由特征 margin（预激活值到截断阈值的距离）决定：在三个 Pythia 模型的五组公开 TopK SAE 对上，将单个特征置零有 0.74 以上概率使下一层特征越过截断阈值，且这些跨越承载了响应的大部分能量。
正文 · 原文
Abstract:Attribution graphs over the features of sparse autoencoders (SAEs) are used to predict what an ablation does and are compared across training seeds. A TopK SAE keeps the largest preactivations at each token, and a feature's margin is its preactivation's distance to the cutoff. On five public TopK SAE pairs from three Pythia models, zeroing one feature moves a next layer feature across the cutoff in 0.74 of cases or more, and the crossings carry most of the response's energy. Over writes that scale a feature's code by three factors, the graph built on the features active beforehand has a median relative error of 0.247 or more. The margin graph adds the first order change to every preactivation and reranks, at the same cost, and errs by at most 0.033. At the median, the crossing features' decoded change is 1.24 times the model's own change in norm or more. Between seeds of one testbed, sparse maps between features explain 0.85 of a graph's energy on its active features, and 0.47 once the other seed's graph keeps only its own. The largest model studied is pythia-160m.
| Comments: | 34 pages, 8 figures |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2611.00606 [cs.LG] |
| (or arXiv:2611.00606v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2611.00606 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Karl Li [view email]
[v1]
Sat, 31 Oct 2026 04:03:14 UTC (199 KB)
来源：arXiv cs.CL · arxiv.org