AI Paper Atlas

Research map / Architectures and efficiency / Spiking networks

Dynamic Token Masking in Spiking Neural Network

Yuetong Fang, Ziqing Wang, Deming Zhou, Hongwei Ren, Shibo Zhou, Renjing Xu

IJCV 2026 · Architectures and efficiency › Spiking networks · approach: Spiking transformers · 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 depth-wise convolution (spatial locality enhancement) at T=1 (ablation)DynMask-T · ImageNet-1Ktop-1 accuracy (%): 66.88 -> 70.47Improved
add depth-wise convolution at T=64 (ablation)DynMask-T · ImageNet-1Ktop-1 accuracy (%): 77.81 -> 80.23Improved
add depth-wise convolution (energy cost vs accuracy, T=8) (ablation)DynMask-T · ImageNet-1Kaccuracy (%) and energy (mJ): 76.82 -> 79.39 accuracy; 7.52 -> 8.55 mJImproved
replace quadratic spiking self-attention with DynMask linear token mixing (energy)MST · ImageNet-1Kenergy (mJ): 380.86 -> 8.55Improved
replace vanilla softmax self-attention (MST T=128) with DynMask at T=8Swin-T (BN) converted SNN · ImageNet-1Ktop-1 accuracy (%): 77.88 -> 79.39Improved
use DynMask under similar membrane-potential-init-free setting (T=48) vs MST with init (T=128)DynMask-T · ImageNet-1Kaccuracy (%) and energy (mJ): 77.79 -> 78.00 accuracy; 99.3 -> 50.61 mJImproved
increase simulation timesteps from 1 to 64DynMask-T · ImageNet-1Ktop-1 accuracy (%): 70.47 -> 80.23Improved
scale model from DynMask-T to DynMask-B at T=64DynMask · ImageNet-1Ktop-1 accuracy (%): 80.23 -> 82.00Improved
replace MST (T=256) with DynMask at T=1Swin-T (BN) converted SNN · CIFAR-10accuracy (%): 97.27 -> 97.75Improved
increase timesteps from 4 to 16DynMask-T · COCO 2017 detectionmAP@0.5: 52.6 -> 60.2; 241.4 -> 405.0 mJImproved
increase timesteps from 4 to 16DynMask-T · ADE20K segmentationMIoU (%): 32.5 -> 38.6Improved
random spike masking of 75% of inputs in spiking self-attention (ablation)Spiking Transformer · CIFAR-10accuracy (%): 97.27 -> 97.29No 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