Research map / Architectures and efficiency / Spiking networks
Training Deep Normalization-Free Spiking Neural Networks with Lateral Inhibition
Peiyu Liu, Jianhao Ding, Zhaofei Yu
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 E-I Init instead of clamped non-negative Kaiming init (ablation) | DeepEISNN · CIFAR-10 (VGG-8) | top-1 accuracy (%): 85.61 -> 87.03 | Improved |
| use E-I constraint with E-I Init instead of unconstrained Kaiming init (ablation) | DeepEISNN · CIFAR-10 (VGG-8) | top-1 accuracy (%): Failed to converge -> 87.03 | Improved |
| use adaptive stabilization of divisive inhibition instead of fixed epsilon 1e-8 to 1e-5 (ablation) | DeepEISNN · CIFAR-10 (VGG-8) | top-1 accuracy (%): Failed to converge/Collapsed -> 87.03 | Improved |
| add gradient scaling 1/d on W_EI (ablation) | DeepEISNN · CIFAR-10 (VGG-8) | top-1 accuracy (%): Collapsed (Epoch 10) -> 87.03 | Improved |
| vary gradient scaling factor from 1/d to 1/sqrt(d) (ablation) | DeepEISNN · CIFAR-10 (VGG-8) | top-1 accuracy (%): 86.88 -> 86.87 | No clear effect |
| use 1/d^2 gradient scaling instead of 1/d (ablation) | DeepEISNN · CIFAR-10 (VGG-8) | top-1 accuracy (%): 86.88 -> 86.33 | Hurt |
| replace vanilla BN with E-I circuit (normalization-free) | ResNet-18 SNN · CIFAR-10 (ResNet-18) | top-1 accuracy (%): 95.37 -> 92.05 | Hurt |
| use E-I circuit with E-I Init/Prop instead of DANN-style E-I network | VGG-16 · CIFAR-10 (VGG-16) | top-1 accuracy (%): 88.54 -> 90.80 | Improved |
| replace BN-based TAB with E-I circuit | SNN · CIFAR10-DVS | top-1 accuracy (%): 76.70 -> 77.66 | Improved |
| replace BN-based TAB with E-I circuit | SNN · CIFAR-100 | top-1 accuracy (%): 75.89 -> 64.95 | Hurt |
Closest papers by meaning
- Training Deep Spiking Neural Networks without Normalization (ICML 2026 · similarity 0.89)
- Intrinsically Stable Spiking Neural Networks: Overcoming the Performance Barrier in the Absence of Batch Normalization (ECCV 2026 · similarity 0.89)
- Direct Training for Spiking Neural Networks: Faster, Larger, Better (AAAI 2019 · similarity 0.89)
- Efficient Training of Deep Spiking Neural Networks with Input-Driven Derivative-Free Updates (NeurIPS 2026 · similarity 0.88)
- NDOT: Neuronal Dynamics-based Online Training for Spiking Neural Networks (ICML 2024 · similarity 0.87)
- Sign Gradient Descent-based Neuronal Dynamics: ANN-to-SNN Conversion Beyond ReLU Network (ICML 2024 · similarity 0.87)
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.
