AI Paper Atlas

Research map / Architectures and efficiency / Spiking networks

Scaling Spike-Driven Transformer With Efficient Spike Firing Approximation Training

Man Yao, Xuerui Qiu, Tianxiang Hu, Jiakui Hu, Yuhong Chou, Keyu Tian, Jianxing Liao, Luziwei Leng, Bo Xu, Guoqi Li

TPAMI 2025 · Architectures and efficiency › Spiking networks · approach: Spiking transformers · 89 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 Spike Firing Approximation (SFA) training (1x4 instead of 4x1) (ablation)Meta-SpikeFormer 10M · ImageNet-1ktop-1 accuracy: 71.3 -> 77.8Improved
add SFA training (ablation)Meta-SpikeFormer 10M · ImageNet-1kpower: 11.9 -> 7.3 mJImproved
add efficient E-SpikeFormer architecture (ablation)Meta-SpikeFormer + SFA · ImageNet-1ktop-1 accuracy: 77.8 -> 78.5Improved
add efficient E-SpikeFormer architecture (ablation)Meta-SpikeFormer + SFA · ImageNet-1kpower: 7.3 -> 3.0 mJImproved
increase inference timestep D from 4 to 8E-SpikeFormer 83M · ImageNet-1ktop-1 accuracy: 83.2 -> 84.0Improved
increase inference timestep D from 4 to 8E-SpikeFormer 173M · ImageNet-1ktop-1 accuracy: 84.7 -> 85.1Improved
increase timestep D from 4 to 8E-SpikeFormer 38.7M · COCO val2017mAP@50: 58.4 -> 58.8No clear effect
increase timestep D from 2 to 4E-SpikeFormer 38.7M · COCO val2017mAP@50: 41.8 -> 58.4Improved
increase timestep D from 2 to 4E-SpikeFormer 11.0M · ADE20KmIoU: 31.9 -> 40.1Improved
increase timestep D from 4 to 8E-SpikeFormer HAR-DVS · HAR-DVSaccuracy: 48.9 -> 49.2No clear effect
use SFA one-timestep training (ablation)Meta-SpikeFormer · ImageNet-1k trainingtraining time: 63 hours -> 12 hoursImproved
apply MIM pre-training with binary spike (vanilla) training when scaling (ablation)SNN · ImageNet-1k scalingaccuracyHurt

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