Research map / Architectures and efficiency / Spiking networks
TP-Spikformer: Token Pruned Spiking Transformer
Wenjie Wei, Xiaolong Zhou, Malu Zhang, Ammar Belatreche, Qian Sun, Yimeng Shan, Dehao Zhang, Zijian Zhou, Zeyu Ma, Yang Yang, Haizhou 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 |
|---|---|---|---|
| use IRToP scoring instead of random token pruning (under IR-Arc) (ablation) | SDT-V1 · ImageNet-1K (no fine-tuning) | accuracy: 60.02% -> 73.78% | Improved |
| use IRToP scoring instead of random token pruning (under IR-Arc) (ablation) | QKFormer · ImageNet-1K (no fine-tuning) | accuracy: 74.45% -> 81.16% | Improved |
| use IRToP scoring instead of random token pruning (under IR-Arc) (ablation) | SDT-V3 · ImageNet-1K (no fine-tuning) | accuracy: 73.15% -> 75.95% | Improved |
| use IR-Arc instead of Drop with random pruning (ablation) | SDT-V1 · ImageNet-1K (no fine-tuning) | accuracy: 59.88% -> 60.02% | No clear effect |
| use IR-Arc instead of Drop (ablation) | QKFormer and SDT-V3 · ImageNet-1K (no fine-tuning) | accuracy | Improved |
| add temporal scorer to spatial scorer (ablation) | SDT-V1 · ImageNet-1K (no fine-tuning) | accuracy: 73.52% -> 73.78% | No clear effect |
| add spatial scorer to temporal scorer (ablation) | QKFormer · ImageNet-1K (no fine-tuning) | accuracy: 79.69% -> 81.16% | Improved |
| add temporal scorer to spatial scorer (ablation) | QKFormer · ImageNet-1K (no fine-tuning) | accuracy: 58.93% -> 81.16% | Improved |
| prune to 20% tokens | Spikformer · CIFAR-10 | accuracy: 95.19 -> 95.12 | No clear effect |
| prune QKFormer to 53% tokens | QKFormer · ImageNet-1K | accuracy: 85.56 -> 82.53 | Mixed |
| prune SDT-V3 to 56% tokens | SDT-V3 · ADE20K | mIoU | No clear effect |
| prune SDT-V3 to 78% tokens with fewer time steps | SDT-V3 · COCO 2017 | mAP@0.5 | Mixed |
Closest papers by meaning
- Towards Efficient Spiking Transformer: a Token Sparsification Framework for Training and Inference Acceleration (ICML 2024 · similarity 0.92)
- STEP: A Unified Spiking Transformer Evaluation Platform for Fair and Reproducible Benchmarking (NeurIPS 2025 · similarity 0.89)
- Spiking Transformer: Introducing Accurate Addition-Only Spiking Self-Attention for Transformer (CVPR 2025 · similarity 0.89)
- TTFSFormer: A TTFS-based Lossless Conversion of Spiking Transformer (ICML 2025 · similarity 0.89)
- Spike2Former: Efficient Spiking Transformer for High-performance Image Segmentation (AAAI 2025 · similarity 0.88)
- Towards efficient deep spiking neural networks construction with spiking activity based pruning (ICML 2024 · similarity 0.88)
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.
