Beyond SAEs: Consistent Concept Discovery with Sparse Coding
Sparse autoencoders (SAEs) explain foundation model representations by decomposing activations into sparse combinations of learned dictionary atoms. However, the learned dictionaries can vary across training runs, limiting the reliability and reproducibility of SAE-based analyses. We propose Consistent Sparse Dictionary Learning (ConSDL), which replaces the standard linear encoder with per-sample…
We haven't written up this one. arXiv cs.AI has the full story — the link below goes straight to it.
