AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Scaling Linear Attention with Sparse State Expansion

Yuqi Pan, Yongqi An, Zheng Li, Yuhong Chou, Rui-Jie Zhu, Xiaohui Wang, Mingxuan Wang, Jinqiao Wang, Guoqi Li

ICLR 2026 · Architectures and efficiency › Attention and transformers · approach: Linear attention approximations

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 softmax instead of SiLU as key feature map (ablation)SSE-n4k1 (600M) · FDA/SWDE/SQuAD recall avgaccuracy: 24.23 -> 31.16Improved
add write-read gate on both KV and Q (ablation)SSE-n4k1 (2B) · Recall avgaccuracy: 51.87 -> 56.63Improved
add write-only gate vs no gate (ablation)SSE-n4k1 (2B) · Recall avgaccuracy: 51.87 -> 50.92No clear effect
add read-only gate vs no gate (ablation)SSE-n4k1 (2B) · Recall avgaccuracy: 51.87 -> 53.96Improved
share attention parameters across partitions (ablation)SSE-n4k1 (600M) · Recall avgaccuracy: 25.95 -> 31.16Improved
increase number of partitions n at constant sparsity ratio (ablation)SSE (2B, 20B tokens) · recallrecall accuracyImproved
increase sparsity ratio k/n beyond about 50% (ablation)SSE (2B, 20B tokens) · recallrecall accuracyNo clear effect
add sparse state expansion (n4k1) to GLA-style linear attentionGLA (2B) · Real-world recall avgaccuracy: 49.29 -> 61.46Improved
add one softmax layer per five linear layers to SSE (hybrid)SSE-n4k1 (2B) · Real-world recall avgaccuracy: 61.46 -> 70.87Improved
use SSE-H hybrid instead of GLA-H2B hybrid · RULER S-NIAH-3 at 8Kaccuracy: 70.8 -> 97.4Improved
use SSE-H instead of Transformer after 2T tokens plus long-context extension2B · MMLUaccuracy: 52.6 -> 54.5Improved
use SSE-H instead of Transformer for reasoning pipeline2B · AIME25avg@32 accuracy: 52.8 -> 50.2No clear effect

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