Research map / Architectures and efficiency / Spiking networks
Spiking Wavelet Transformer
Yuetong Fang, Ziqing Wang, Lingfeng Zhang, Jiahang Cao, Honglei Chen, Renjing Xu
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 |
|---|---|---|---|
| Replace spiking self-attention with the Frequency-Aware Token Mixer | Spiking Transformer · ImageNet | top-1 accuracy (%): 72.46 -> 74.98 | Improved |
| Add spiking wavelet transform branch (ablation) | SWformer · CIFAR-100 | accuracy (%): 78.89 -> 79.31 | Improved |
| Add spiking wavelet transform branch (ablation) | SWformer · CIFAR10-DVS | accuracy (%): 80.9 -> 83.9 | Improved |
| Add negative spike dynamics (ternary spikes) in the wavelet transform (ablation) | SWformer · CIFAR-100 | accuracy (%): 78.64 -> 79.31 | Improved |
| Add spiking Haar inverse transform (ablation) | SWformer · CIFAR-100 | accuracy (%): 78.79 -> 79.31 | Improved |
| Add 1D max-pooling at the start of FATM (ablation) | SWformer · CIFAR10-DVS | accuracy (%): 81.8 -> 83.9 | Improved |
| Mask DC (low-frequency) components (ablation) | SWformer · CIFAR-100 / CIFAR10-DVS | accuracy (%): 79.31 -> 79.34; 83.9 -> 84.0 | No clear effect |
| Increase weight splitting blocks from 2 to 4 (ablation) | SWformer · ImageNet | parameters and accuracy: 21.8M -> 18.46M params, 74.98 -> 74.62 (Vth=0.5) | Improved |
| Use standard non-spiking wavelet instead of spiking Haar (ablation) | SWformer* · ImageNet | top-1 accuracy (%): 75.18 -> 75.26 (B=2, Transformer-8-512) | Improved |
| Increase firing threshold of spiking frequency representation (ablation) | SWformer (Block=4) · ImageNet | power: 58.4% to 60.1% reduction | Improved |
| Use SWformer on neuromorphic data | N-Caltech101, N-Cars · neuromorphic datasets | accuracy (%): NDA 83.70 -> 88.45; 91.90 -> 96.32 | Improved |
Closest papers by meaning
- Spiking Transformer: Introducing Accurate Addition-Only Spiking Self-Attention for Transformer (CVPR 2025 · similarity 0.88)
- Spiking Transformer with Spatial-Temporal Attention (CVPR 2025 · similarity 0.87)
- Spiking Neural Networks Need High-Frequency Information (NeurIPS 2025 · similarity 0.87)
- Spike-driven Transformer V2: Meta Spiking Neural Network Architecture Inspiring the Design of Next-generation Neuromorphic Chips (ICLR 2024 · similarity 0.87)
- FSTA-SNN:Frequency-Based Spatial-Temporal Attention Module for Spiking Neural Networks (AAAI 2025 · similarity 0.86)
- SpikingResformer: Bridging ResNet and Vision Transformer in Spiking Neural Networks (CVPR 2024 · similarity 0.86)
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.
