Research map / Architectures and efficiency / Spiking networks
Rethinking Attention in Spiking Transformers: Overcoming Density Bias with Set Similarity
JinGyo Lim, Seung Gyu Jeong, Seong-Eun Kim
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 SADA then SDA with 5:5 ratio (full model) instead of SDSA only (ablation) | DiceFormer-10-L · AudioSet-20k | mAP: 0.126 -> 0.161 | Improved |
| use full SADA then SDA instead of SSA only (ablation) | DiceFormer-10-L · AudioSet-20k | mAP: 0.142 -> 0.161 | Improved |
| order SADA before SDA instead of SDA before SADA (ablation) | DiceFormer-10-L · AudioSet-20k | mAP: 0.157 -> 0.161 | Improved |
| combine SADA and SDA instead of SADA only (ablation) | DiceFormer-10-L · AudioSet-20k | mAP: 0.157 -> 0.161 | Improved |
| SADA:SDA ratio 5:5 instead of 6:4 (ablation) | DiceFormer-10-L · AudioSet-20k | mAP: 0.160 -> 0.161 | Improved |
| SADA:SDA ratio 5:5 instead of 4:6 (ablation) | DiceFormer-10-L · AudioSet-20k | mAP: 0.155 -> 0.161 | Improved |
| binarize the attention map (ablation) | DiceFormer-10-L · AudioSet-20k | mAP: 0.160 -> 0.161 | No clear effect |
| use Lin-SDA instead of SDA (ablation) | DiceFormer-10-L · AudioSet-20k | mAP: 0.161 -> 0.159 | No clear effect |
| add SDA to Spike-driven Transformer | Spike-driven Transformer · AudioSet-20k | mAP: 0.130 -> 0.142 | Improved |
| add SDA to Spike-driven Transformer V2 | Spike-driven Transformer V2 · AudioSet-20k | mAP: 0.138 -> 0.144 | Improved |
| add SDA to QKFormer | QKFormer · AudioSet-20k | mAP: 0.147 -> 0.149 | Improved |
| add SDA to Spike-driven Transformer | Spike-driven Transformer · CIFAR-100 | top-1 accuracy (%): 78.40 -> 78.99 | Improved |
Closest papers by meaning
- STEP: A Unified Spiking Transformer Evaluation Platform for Fair and Reproducible Benchmarking (NeurIPS 2025 · similarity 0.91)
- Rethinking Spiking Self-Attention Mechanism: Implementing a-XNOR Similarity Calculation in Spiking Transformers (CVPR 2025 · similarity 0.90)
- Spiking Transformer with Spatial-Temporal Attention (CVPR 2025 · similarity 0.90)
- Spiking Transformer: Introducing Accurate Addition-Only Spiking Self-Attention for Transformer (CVPR 2025 · similarity 0.90)
- Towards Efficient Spiking Transformer: a Token Sparsification Framework for Training and Inference Acceleration (ICML 2024 · similarity 0.88)
- SpikingResformer: Bridging ResNet and Vision Transformer in Spiking Neural Networks (CVPR 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.
