AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

On the Limits of Sparse Autoencoders: A Theoretical Framework and Reweighted Remedy

Jingyi Cui, Qi Zhang, Yifei Wang, Yisen Wang

ICLR 2026 · Architectures and efficiency › RNNs and normalization · approach: Sparse dictionary representations · 9 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
reweight SAE reconstruction loss toward high-variance (monosemantic) dimensions, alpha=1TopK SAE on Pythia-160M · Auto-interpretability, layers 0-11auto-interpretability score (%): average +3.8% gain; layer 3: 77.8 -> 84.6Improved
use alpha=1 instead of alpha=0.5 in the weighting exponent (ablation)Weighted SAE on Pythia-160M · Auto-interpretability, layers 9-11auto-interpretability score (%): layer 10: 76.1 (alpha=0.5) vs 72.4 (alpha=1); layer 3: 77.9 vs 84.6Mixed
add reweighted loss to SAE with grouped Top-1 TopK variantSAE on Pythia-160M · Auto-interpretability, layers 9-11auto-interpretability score (%): +2.5, +1.9, +2.7 (layers 9, 10, 11)Improved
add reweighted loss to SAESAE on Llama-3-8B · Auto-interpretability, layers 8-32auto-interpretability score (%): +4.5, +2.8, +3.8, +2.2 (layers 8, 16, 24, 32)Improved
weight SAE loss by semantic consistency, alpha>0TopK SAE on NCL ResNet-18 (ImageNet-100) · Semantic consistencysemantic consistency (%)Improved
vary alpha in {0.5,...,2.5} versus alpha=0 (plain SAE) (ablation)WSAE on NCL ResNet-18 · Semantic consistency sensitivitysemantic consistencyImproved
increase ground-truth feature sparsity in synthetic data (ablation)standard SAE · Synthetic toy superposition (n=200, n_p=20)avg number of ground-truth features activated per latentImproved
reweight loss in WSAE on synthetic data at low sparsitySAE · Synthetic toy superpositionground truth reconstruction errorImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy