ICML 2026poster0 citations

On the Relationship Between Activation Outliers and Feature Death in Sparse Autoencoders

Elana Simon, Etowah Adams, James Zou

Abstract

Sparse autoencoders (SAEs) decompose neural network activations into interpretable features, but many features never activate- a problem called feature death. Death rates vary dramatically across models: near-zero on GPT-2, over 70\% on AlphaFold3 with identical SAE configurations. Why? We find that dimension-level activation outliers (dimensions where mean magnitude is large relative to per-token variation) shift pre-activations at initialization, making feature fate depend on weight-outlier alignment rather than input content. We derive $\gamma = \|\boldsymbol{\mu}\|/\|\boldsymbol{\sigma}\|$ from this mechanism; it predicts initial death rates (Spearman $\rho > 0.9$) across 275 model-layer combinations spanning language, vision, and protein models. This creates two death pathways; we trace their recovery mechanisms and find one resolves naturally while the other bottlenecks on the SAE slowly learning to mean-center. Initializing the SAE to mean-center from the start eliminates this outlier-induced death, confirming the mechanism.

VisionHealthcare
BibTeX
@inproceedings{
simon2026on,
title={On the Relationship Between Activation Outliers and Feature Death in Sparse Autoencoders},
author={Elana Simon and Etowah Adams and James Zou},
booktitle={Forty-third International Conference on Machine Learning},
year={2026},
url={https://openreview.net/forum?id=hdak3I0DYY}
}