研究：SAE 归因图结果取决于特征的边际距离
Sparse Autoencoder Attribution Graphs Are Decided at the Margin
研究 SAE 可解释性的朋友可以看看，论文指出传统归因图在特征靠近激活阈值时预测会失效，边际图方法误差能从 0.247 降到 0.033。
arXiv 论文分析稀疏自编码器（SAE）特征上的归因图如何预测消融效果。基于 3 个 Pythia 模型的 5 对公开 TopK SAE，发现将一个特征置零时，有 0.74 以上概率使下一层某个特征跨越激活阈值，且这些跨越承载了响应的主要能量。传统归因图在按 3 倍缩放特征的写入操作上中位相对误差达 0.247 以上，而新提出的边际图方法将一阶变化加入每个 preactivation 后重排序，误差降至 0.033 以内。在跨训练种子比较中，稀疏映射可解释一个图在活跃特征上 0.85 的能量，只保留自身特征时降到 0.47。
Sparse Autoencoder Attribution Graphs Are Decided at the Margin
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.