Research map / Architectures and efficiency / Spiking networks
Optimal Conversion of Conventional Artificial Neural Networks to Spiking Neural Networks
Shikuang Deng, Shi Gu
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 threshold ReLU rather than regular ReLU in the source ANN (ablation) | ANN-to-SNN conversion · CIFAR-10, CIFAR-Net; simulation length 400-600 | accuracy: 90.16±0.03 -> 90.61±0.01 | Improved |
| Use threshold ReLU rather than regular ReLU in the source ANN (ablation) | ANN-to-SNN conversion · CIFAR-100, ResNet-20; simulation length 400-600 | accuracy: 66.74±0.04 -> 69.82±0.03 | Improved |
| Add shift to threshold ReLU (ablation) | ANN-to-SNN conversion · CIFAR-10, CIFAR-Net; simulation length 400-600 | accuracy: 90.61±0.01 -> 90.37±0.02 | Hurt |
| Add shift to threshold ReLU (ablation) | ANN-to-SNN conversion · CIFAR-10, VGG-16; simulation length 400-600 | accuracy: 92.26±0.01 -> 92.26±0.01 | No clear effect |
| Add shift to threshold ReLU (ablation) | ANN-to-SNN conversion · CIFAR-100, ResNet-20; simulation length 400-600 | accuracy: 69.82±0.03 -> 69.81±0.02 | No clear effect |
| Combine threshold ReLU with shift instead of using threshold alone (ablation) | ANN-to-SNN conversion · CIFAR-100, VGG-16; simulation length 128 | conversion loss: 0.21%(0.08%) -> 0.15%(0.02%) | Improved |
| Combine threshold ReLU with shift instead of using threshold alone (ablation) | ANN-to-SNN conversion · CIFAR-10, ResNet-20; simulation length 32 | conversion loss: 19.72%(21.01%) -> -0.98%(0.31%) | Improved |
| Apply threshold ReLU plus shift instead of regular ReLU | ANN-to-SNN conversion · CIFAR-10, VGG-16; simulation length 32 | conversion loss: RMP: 33.33%; Our Work-TS: -0.2%(0.05%) | Improved |
| Apply threshold ReLU plus shift instead of regular ReLU | ANN-to-SNN conversion · CIFAR-100, ResNet-20; simulation length 32 | conversion loss: RMP: 41.08%; Our Work-TS: -1.32%(1.4%) | Improved |
| Use threshold ReLU rather than regular ReLU in the source RNN (ablation) | RNN-to-SNN conversion · Movie Reviews validation set | accuracy: 0.5100 -> 0.5430 | Improved |
| Convert the threshold-ReLU RNN rather than directly train an SNN with surrogate gradient | Sentiment analysis SNN · Movie Reviews validation set | accuracy: 0.5106 -> 0.5430 | Improved |
| Use threshold ReLU and shift rather than threshold ReLU alone (ablation) | ANN-to-SNN conversion · ResNet-20, CIFAR-100; simulation length 16 | conversion loss: 3.35%(6.07%) -> -1.32%(1.40%) | Improved |
Closest papers by meaning
- Faster and Stronger: When ANN-SNN Conversion Meets Parallel Spiking Calculation (ICML 2025 · similarity 0.93)
- Inference-Scale Complexity in ANN-SNN Conversion for High-Performance and Low-Power Applications (CVPR 2025 · similarity 0.92)
- Efficient ANN-SNN Conversion with Error Compensation Learning (ICML 2025 · similarity 0.92)
- Differential Coding for Training-Free ANN-to-SNN Conversion (ICML 2025 · similarity 0.91)
- Adaptive Calibration: A Unified Conversion Framework of Spiking Neural Networks (AAAI 2025 · similarity 0.91)
- Temporal Misalignment in ANN-SNN Conversion and its Mitigation via Probabilistic Spiking Neurons (ICML 2025 · similarity 0.91)
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.
