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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add Spike Firing Approximation (SFA) training (1x4 instead of 4x1) (ablation) | Meta-SpikeFormer 10M · ImageNet-1k | top-1 accuracy: 71.3 -> 77.8 | Improved |
| add SFA training (ablation) | Meta-SpikeFormer 10M · ImageNet-1k | power: 11.9 -> 7.3 mJ | Improved |
| add efficient E-SpikeFormer architecture (ablation) | Meta-SpikeFormer + SFA · ImageNet-1k | top-1 accuracy: 77.8 -> 78.5 | Improved |
| add efficient E-SpikeFormer architecture (ablation) | Meta-SpikeFormer + SFA · ImageNet-1k | power: 7.3 -> 3.0 mJ | Improved |
| increase inference timestep D from 4 to 8 | E-SpikeFormer 83M · ImageNet-1k | top-1 accuracy: 83.2 -> 84.0 | Improved |
| increase inference timestep D from 4 to 8 | E-SpikeFormer 173M · ImageNet-1k | top-1 accuracy: 84.7 -> 85.1 | Improved |
| increase timestep D from 4 to 8 | E-SpikeFormer 38.7M · COCO val2017 | mAP@50: 58.4 -> 58.8 | No clear effect |
| increase timestep D from 2 to 4 | E-SpikeFormer 38.7M · COCO val2017 | mAP@50: 41.8 -> 58.4 | Improved |
| increase timestep D from 2 to 4 | E-SpikeFormer 11.0M · ADE20K | mIoU: 31.9 -> 40.1 | Improved |
| increase timestep D from 4 to 8 | E-SpikeFormer HAR-DVS · HAR-DVS | accuracy: 48.9 -> 49.2 | No clear effect |
| use SFA one-timestep training (ablation) | Meta-SpikeFormer · ImageNet-1k training | training time: 63 hours -> 12 hours | Improved |
| apply MIM pre-training with binary spike (vanilla) training when scaling (ablation) | SNN · ImageNet-1k scaling | accuracy | Hurt |
Closest papers by meaning
- Spike-driven Transformer V2: Meta Spiking Neural Network Architecture Inspiring the Design of Next-generation Neuromorphic Chips (ICLR 2024 · similarity 0.93)
- Adaptive Calibration: A Unified Conversion Framework of Spiking Neural Networks (AAAI 2025 · similarity 0.92)
- Inference-Scale Complexity in ANN-SNN Conversion for High-Performance and Low-Power Applications (CVPR 2025 · similarity 0.92)
- Spiking Transformer: Introducing Accurate Addition-Only Spiking Self-Attention for Transformer (CVPR 2025 · similarity 0.92)
- SpikedAttention: Training-Free and Fully Spike-Driven Transformer-to-SNN Conversion with Winner-Oriented Spike Shift for Softmax Operation (NeurIPS 2024 · similarity 0.91)
- EnOF-SNN: Training Accurate Spiking Neural Networks via Enhancing the Output Feature (NeurIPS 2024 · similarity 0.91)
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.
