AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Taming Polysemanticity in LLMs: Theory-Grounded Feature Recovery via Sparse Autoencoders

Siyu Chen, Heejune Sheen, Xuyuan Xiong, Tianhao Wang, Zhuoran Yang

ICLR 2026 · Architectures and efficiency › RNNs and normalization · approach: Sparse dictionary representations

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
use TopK with K mismatched to data sparsity (K=20 to K=50) (ablation)TopK SAE · Imbalanced synthetic (n=128, d=42, M=8192)FRR (tau_align>=0.9): 98.4% -> 23.4%Hurt
use GBA with full frequency coverage instead of TopKSAE training · Imbalanced synthetic (n=128, d=42, M=8192)FRR (tau_align>=0.9): 23.4% -> 100.0%Improved
misspecify GBA frequency coverage (ablation)GBA · Imbalanced synthetic (n=128, d=42, M=8192)FRR (tau_align>=0.8): 100.0% -> 38.3%Hurt
use GBA instead of TopK for cross-seed consistencySAE training · Qwen2.5-1.5B Github L26, top-10% neuronsFraction MCS>0.9: 0.0317 -> 0.0366Improved
add neuron grouping (GBA vs single-group BA) (ablation)Bias adaptation · Qwen2.5-1.5B Pile reconstruction vs sparsityReconstruction loss at equal sparsity: significantly outperforming both the L1 penalty method and the non-grouped variant BAImproved
compare GBA with TopK on reconstruction-sparsity frontierGBA · Qwen2.5-1.5B Pile reconstruction vs sparsityReconstruction loss: comparable to TopK's best curveNo clear effect
use GBA instead of TopK on SAEBench explained varianceSAE training · SAEBench (Gemma2-2B, layer 12, L0 300-400)Explained variance: 0.859 -> 0.902Improved
use GBA instead of TopK on absorption scoreSAE training · SAEBench (Gemma2-2B, layer 12, L0 300-400)Absorption score: 0.0269 -> 0.0041Improved
use GBA instead of TopK on alive fractionSAE training · SAEBench (Gemma2-2B, layer 12, L0 300-400)Alive fraction: 0.872 -> 0.970Improved
use GBA instead of TopK on TPP and SCRSAE training · SAEBench (Gemma2-2B, layer 12, L0 300-400)TPP / SCR: 0.213 -> 0.184 (TPP); 0.332 -> 0.296 (SCR)Hurt
increase number of groups K to 10 or more (ablation)GBA · Github Layer 26Reconstruction loss and sparsity: stabilize when K >= 10No clear effect
increase highest target frequency HTF toward 0.5 (ablation)GBA · Github Layer 26Reconstruction loss: HTF 0.05 gives higher reconstruction loss; stabilizes at HTF = 0.5Improved

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