Research map / Architectures and efficiency / Spiking networks
SpiLiFormer: Enhancing Spiking Transformers with Lateral Inhibition
Zeqi Zheng, Yanchen Huang, Yingchao Yu, Zizheng Zhu, Junfeng Tang, Zhaofei Yu, Yaochu Jin
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 FF-LiDiff attention (ablation) | SpiLiFormer · CIFAR-10 | top-1 accuracy (%): 96.32 -> 96.63 | Improved |
| add FB-LiDiff attention (feedback second forward pass) (ablation) | SpiLiFormer · CIFAR-10 | top-1 accuracy (%): 96.01 -> 96.63 | Improved |
| add FF-LiDiff attention (ablation) | SpiLiFormer · CIFAR10-DVS | top-1 accuracy (%): 85.1 -> 86.7 | Improved |
| add FB-LiDiff attention (feedback second forward pass) (ablation) | SpiLiFormer · CIFAR10-DVS | top-1 accuracy (%): 83.7 -> 86.7 | Improved |
| set loss weight alpha to 0.5 rather than 0.7 (ablation) | SpiLiFormer · CIFAR-10 | top-1 accuracy (%): 96.16 -> 96.63 | Improved |
| set loss weight alpha to 0.5 rather than 0.7 (ablation) | SpiLiFormer · CIFAR10-DVS | top-1 accuracy (%): 85.6 -> 86.7 | Improved |
| set loss weight alpha to 0.4 instead of 0.5 (ablation) | SpiLiFormer · CIFAR10-DVS | top-1 accuracy (%): 86.4 -> 86.7 | No clear effect |
| add FB-LiDiff attention (extra forward pass) (ablation) | SpiLiFormer · CIFAR-10 inference latency | inference time per sample (ms): 0.7657 -> 0.9322 (+21.7%) | Hurt |
| add FB-LiDiff attention (extra forward pass) (ablation) | SpiLiFormer · ImageNet-1K inference latency | inference time per sample (ms): 58.9901 -> 66.3027 (+12.4%) | Hurt |
| add FB-LiDiff attention (extra forward pass) (ablation) | SpiLiFormer · CIFAR10-DVS inference latency | inference time per sample (ms): 9.9433 -> 10.0657 (+1.2%) | Hurt |
| replace QKFormer attention with lateral-inhibition attention | QKFormer · ImageNet-1K (224, T=4) | top-1 accuracy (%): 84.22 -> 85.82 | Improved |
| replace QKFormer attention with lateral-inhibition attention | QKFormer · CIFAR10-DVS | top-1 accuracy (%): 84.0 -> 86.7 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Spikformer-8-768 — Spikformer: When Spiking Neural Network Meets Transformer (ICLR 2023) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Spiking Transformer: Introducing Accurate Addition-Only Spiking Self-Attention for Transformer (CVPR 2025 · similarity 0.91)
- SpikingResformer: Bridging ResNet and Vision Transformer in Spiking Neural Networks (CVPR 2024 · similarity 0.90)
- STEP: A Unified Spiking Transformer Evaluation Platform for Fair and Reproducible Benchmarking (NeurIPS 2025 · similarity 0.90)
- Spiking Transformer with Spatial-Temporal Attention (CVPR 2025 · similarity 0.89)
- Spiking Vision Transformer with Saccadic Attention (ICLR 2025 · similarity 0.88)
- Toward Relative Positional Encoding in Spiking Transformers (NeurIPS 2025 · 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.
