Unstable Features, Reproducible Subspaces: Understanding Seed Dependence in Sparse Autoencoders
Sparse autoencoders (SAEs) are widely used to interpret neural network representations, but their utility depends on whether the learned features are reproducible across training runs. We study this question through \emph{feature stability}: for each SAE feature, we estimate the probability that a similar feature reappears in an independently trained SAE. This yields a scalable per-feature signal that separates stable from unstable features. In a large-scale study across seeds, models, layers, dictionary sizes, and SAE variants, we find a pronounced functional asymmetry: stable features carry most of the reconstruction- and prediction-relevant signal, while unstable features have weak mar...
| Authors | Gleb Gerasimov, Timofei Rusalev, Nikita Balagansky et al. |
|---|---|
| Primary category | cs.LG |
| Published | 10 June 2026 |
| Last updated at source | 10 June 2026 |
Silytics shows public metadata from arXiv and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.