Research map / Architectures and efficiency / Spiking networks
Training Deep Spiking Neural Networks without Normalization
Xinyu Shi, Zhaofei Yu
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 SpikeInit weight initialization (firing-rate response curve) (ablation) | normalization-free SNN · CIFAR-10 VGG-11 / ResNet-18 | accuracy: fails to converge -> 95.29 (VGG-11); fails -> 95.53 (ResNet-18) | Improved |
| add surrogate gradient shape parameter initialization (ablation) | SpikeInit · CIFAR-10 VGG-11 | accuracy: 78.96 -> 95.29 | Improved |
| add surrogate gradient shape parameter initialization (ablation) | SpikeInit · CIFAR-10 ResNet-18 | accuracy: 91.05 -> 95.53 | Improved |
| add adaptive surrogate gradient (ablation) | SpikeInit · CIFAR-10 VGG-11 | accuracy: 94.43 -> 95.29 | Improved |
| add adaptive surrogate gradient (ablation) | SpikeInit · CIFAR-10 ResNet-18 | accuracy: 94.91 -> 95.53 | Improved |
| replace Micheli et al. initialization with SpikeInit | normalization-free SNN · CIFAR-10 VGG-11/VGG-16 | accuracy: fails to converge -> 95.29 (VGG-11), 95.77 (VGG-16) | Improved |
| replace Micheli et al. initialization with SpikeInit | MS ResNet-34 · CIFAR-10 | accuracy: 91.83 -> 96.20 | Improved |
| remove normalization and use SpikeInit instead of TEBN | MS ResNet-34 · CIFAR-10 | accuracy: 95.87 -> 96.20 | Improved |
| remove normalization and use SpikeInit instead of TEBN | MS ResNet-34 · ImageNet (4 time steps) | accuracy: 68.28 -> 68.81 | Improved |
| replace LayerNorm with SpikeInit in spiking transformer | spiking transformer 4-256 · IWSLT17 German-English | BLEU: 29.50 -> 29.98 | Improved |
| replace BatchNorm (tdBN) with SpikeInit in spiking transformer | spiking transformer 4-256 · IWSLT17 German-English | BLEU: 28.14 -> 29.98 | Improved |
| increase depth to 1000 layers with SpikeInit vs tdBN baseline | MS-ResNet · CIFAR-10 depth scaling | accuracy | Improved |
Closest papers by meaning
- Efficiently Training Time-to-First-Spike Spiking Neural Networks from Scratch (ICML 2026 · similarity 0.91)
- Direct Training for Spiking Neural Networks: Faster, Larger, Better (AAAI 2019 · similarity 0.91)
- Going Deeper With Directly-Trained Larger Spiking Neural Networks (AAAI 2021 · similarity 0.90)
- Temporal Effective Batch Normalization in Spiking Neural Networks (NeurIPS 2022 · similarity 0.90)
- Take A Shortcut Back: Mitigating the Gradient Vanishing for Training Spiking Neural Networks (NeurIPS 2024 · similarity 0.89)
- Efficient Logit-based Knowledge Distillation of Deep Spiking Neural Networks for Full-Range Timestep Deployment (ICML 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.
