AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Improving Sparse Autoencoder with Dynamic Attention

Dongsheng Wang, Jinsen Zhang, Dawei Su, Hui Huang

CVPR 2026 · Architectures and efficiency › RNNs and normalization · approach: Sparse dictionary representations · 2 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
add sparsemax attention (replace ReLU) in transformer SAE (ablation)Transformer SAE · ImageNet zero-shot, on_1accuracy: 3.86 -> 10.93Improved
add sparsemax attention (replace ReLU) in transformer SAE (ablation)Transformer SAE · ImageNet zero-shot, on_50accuracy: 36.33 -> 59.95Improved
add transformer cross-attention architecture (replace MLP) with sparsemax (ablation)Sparsemax SAE · ImageNet zero-shot, on_1accuracy: 7.91 -> 10.93Improved
add transformer cross-attention architecture (replace MLP) with sparsemax (ablation)Sparsemax SAE · ImageNet zero-shot, on_50accuracy: 55.32 -> 59.95Improved
add transformer architecture with ReLU to ReLU SAE (ablation)ReLU SAE · ImageNet zero-shot, on_10accuracy: 22.17 -> 24.08Improved
add sparsemax to MLP SAE (vs ReLU SAE) (ablation)MLP SAE · ImageNet zero-shot, on_10accuracy: 22.17 -> 39.73Improved
set K by the average sparsemax activation count (K=24 instead of 32) (ablation)BatchTopK · ImageNet zero-shot, on_10accuracy: 48.99 -> 49.52Improved
set K=24 instead of 32 (ablation)TopK · ImageNet zero-shot, on_1accuracy: 8.64 -> 0.99Hurt
set K=24 instead of 32 (ablation)TopK · ImageNet zero-shot, on_10accuracy: 30.21 -> 42.88Improved
use Sparsemax SAE instead of BatchTopK (K=24)SAE · OpenWebText NMSE (M=3072)NMSE: 0.014 -> 0.005Improved
use Sparsemax SAE instead of BatchTopKSAE · OpenWebText CE degradation (M=3072)CE degradation: 0.196 -> 0.031Improved

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