AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Neural Attention Search Linear: Towards Adaptive Token-Level Hybrid Attention Models

Difan Deng, Andreas Bentzen Winje, Lukas Fehring, Marius Lindauer

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Linear attention approximations · 0 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 softmax attention in inner-chunk computation alongside GDN (ablation)NAtS-L Hybrid · Retrieval tasks avgavg accuracy: 26.98 -> 33.49Improved
add GDN in inner-chunk computation alongside softmax attention (ablation)NAtS-L Hybrid · Retrieval tasks avgavg accuracy: 30.87 -> 33.49Improved
add decay of hidden states for softmax chunks (ablation)NAtS-L Hybrid · Retrieval tasks avgavg accuracy: 32.24 -> 33.49Improved
add separate normalization of each attention output before summing (ablation)NAtS-L Hybrid · Retrieval tasks avgavg accuracy: 29.59 -> 33.49Improved
add output weights for combining attention operations (ablation)NAtS-L Hybrid · Retrieval tasks avgavg accuracy: 32.34 -> 33.49Improved
use softmax instead of sigmoid for output weights (ablation)NAtS-L Hybrid · Retrieval tasks avgavg accuracy: 30.02 -> 33.49Improved
use softmax instead of linear output weights (ablation)NAtS-L Hybrid · Retrieval tasks avgavg accuracy: 30.57 -> 33.49Improved
compute output weights from q instead of input X (ablation)NAtS-L Hybrid · Retrieval tasks avgavg accuracy: 27.64 -> 33.49Improved
use chunk size 64 instead of 128 (ablation)NAtS-L Hybrid · Retrieval tasks avgavg accuracy: 27.85 -> 33.49Improved
insert NAtS-L layers into GDN layers (hybrid) versus NAtS-L onlyNAtS-L · Retrieval tasks avg (800M)avg accuracy: 39.75 -> 45.07Improved
NAtS-L Hybrid versus layer-wise GDN HybridGDN Hybrid · Retrieval tasks avg (800M)avg accuracy: 37.30 -> 45.07Improved
NAtS-L Hybrid versus GDN Hybrid at longer contextGDN Hybrid · RULER 16kmean score: 0.00 -> 0.21Improved

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