AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Linearizing Vision Transformer with Test-Time Training

Yining Li, Dongchen Han, Zeyu Liu, Hanyi Wang, Yulin Wang, Gao Huang

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Attention head and block redesign · 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
inherit both attention and MLP weights instead of MLP only (ablation)T5 (DiT-S/2) · ImageNet 256 generationFID: 89.71 -> 68.52Improved
inherit both attention and MLP weights instead of attention only (ablation)T5 (DiT-S/2) · ImageNet 256 generationFID: 93.18 -> 68.52Improved
add InstanceNorm on keys (vs none) (ablation)TTT linearization · DeiT-Tiny ImageNetTop-1 (%): 0.37 -> 71.19Improved
use InstanceNorm instead of LayerNorm on keys (ablation)TTT linearization · DeiT-Tiny ImageNetTop-1 (%): 57.25 -> 71.19Improved
add mean subtraction to key InstanceNorm (ablation)TTT linearization · DeiT-Tiny ImageNetTop-1 (%): 51.43 -> 71.19Improved
add division-by-std to key InstanceNorm (ablation)TTT linearization · DeiT-Tiny ImageNetTop-1 (%): 71.15 -> 71.19No clear effect
add depthwise convolution on queries and keys (vs none) (ablation)TTT · DeiT-Tiny ImageNetTop-1 (%): 69.25 -> 71.19Improved
use DWC on QK instead of DWC on values (ablation)TTT · DeiT-Tiny ImageNetTop-1 (%): 70.47 -> 71.19Improved
add Neighborhood Attention (window 5) to T5 (ablation)T5 · DeiT-Tiny ImageNetTop-1 (%): 71.19 -> 72.06Improved
add NAT (ablation)T5 (DiT-S/2) · ImageNet 256 generationFID: 72.98 -> 68.52Improved
use 2x learning rate for TTT-specific params (ablation)T5 (DiT-S/2) · ImageNet 256 generationFID: 68.84 -> 68.52Improved
raise TTT-parameter learning rate multiplier from 1x to 20x (ablation)T5 · DeiT-Tiny ImageNetTop-1 (%): 68.14 -> 71.19Improved

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