Research map / Architectures and efficiency / Spiking networks
SpikePack: Enhanced Information Flow in Spiking Neural Networks with High Hardware Compatibility
Guobin Shen, Jindong Li, Tenglong Li, Dongcheng Zhao, Yi Zeng
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 LIF neurons with SpikePack in direct training | ResNet-34 (T=4) · ImageNet-1k | top-1 accuracy (%): SpikePack 73.4 vs T-RevSNN 73.2 / TET 68.0 / SEW ResNet 67.0 on ResNet-18/34 baselines | Improved |
| increase time steps from 4 to 8 with SpikePack (ablation) | Spikeformer-8-512 · ImageNet-1k | top-1 accuracy (%): 78.4 -> 80.1 | Improved |
| increase time steps from 2 to 8 with SpikePack (ablation) | Spikeformer-8-512 · ImageNet-1k | training memory (MB/img): 150.5 -> 150.5 | No clear effect |
| use SpikePack for ANN-to-SNN conversion instead of other conversion methods | ResNet-34 / ViT-L/14 · ImageNet-1k conversion | top-1 accuracy (%): 77.92 (ResNet-34), 88.27 (ViT-L/14) at few time steps | Improved |
| increase time steps for SpikePack on neuromorphic data (ablation) | SpikePack · CIFAR10-DVS | accuracy (%): 68.1 -> 84.6 | Improved |
| increase time steps from 10 to 16 for SpikePack (ablation) | SpikePack · DVS-Gesture | accuracy (%): 97.4 -> 97.4 | No clear effect |
| increase time steps from 6 to 10 for SpikePack detector (ablation) | SpikePack w/ DINO-r50 · COCO 2017 val | mAP@50:95 (%): 48.5 -> 50.1 | Improved |
| increase time steps from 6 to 10 for SpikePack segmentation (ablation) | SpikePack w/ Segformer-b2 · ADE20K | mIoU (%): 42.8 -> 45.6 | Improved |
| set membrane time constant to 4.0 instead of 2.0 (ablation) | SpikePack ResNet-34 · ImageNet-1k | error rate (%) | Improved |
| replace LIF with SpikePack on FPGA neuromorphic processor | ResNet-34 · FPGA simulation | latency: 29.1 ms -> 23ms | Improved |
| replace LIF with SpikePack on FPGA neuromorphic processor | ResNet-34 · FPGA simulation | energy: 23.8 mJ -> 18.6mJ | Improved |
Closest papers by meaning
- Spike-driven Transformer V2: Meta Spiking Neural Network Architecture Inspiring the Design of Next-generation Neuromorphic Chips (ICLR 2024 · similarity 0.90)
- Adaptive Calibration: A Unified Conversion Framework of Spiking Neural Networks (AAAI 2025 · similarity 0.88)
- Are Conventional SNNs Really Efficient? A Perspective from Network Quantization (CVPR 2024 · similarity 0.88)
- TS-SNN: Temporal Shift Module for Spiking Neural Networks (ICML 2025 · similarity 0.88)
- Faster and Stronger: When ANN-SNN Conversion Meets Parallel Spiking Calculation (ICML 2025 · similarity 0.88)
- S$^2$NN: Sub-bit Spiking Neural Networks (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.
