Research map / Architectures and efficiency / Spiking networks
Towards Ultra Low Latency Spiking Neural Networks for Vision and Sequential Tasks Using Temporal Pruning
Sayeed Shafayet Chowdhury, Nitin Rathi, Kaushik Roy
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 gradual temporal pruning instead of direct conversion to T1 (ablation) | SNN VGG6 · CIFAR10 | top-1 accuracy (%): 88.64 -> 89.10 | Improved |
| add gradual temporal pruning instead of direct T5 to T1 transition (ablation) | SNN VGG16 · CIFAR10 | training convergence | Improved |
| train ANN with batch-norm before conversion (ablation) | SNN T1 VGG16 · CIFAR10 | top-1 accuracy (%): 91.03 -> 93.05 | Improved |
| train ANN with batch-norm before conversion (ablation) | SNN T1 VGG16 · CIFAR100 | top-1 accuracy (%): 67.76 -> 70.15 | Improved |
| train ANN with batch-norm before conversion (ablation) | SNN T5 VGG16 · CIFAR10 | top-1 accuracy (%): 92.15 -> 93.90 | Improved |
| include all intermediate timesteps rather than skipping to T3 (ablation) | SNN T1 VGG16 · CIFAR10 | top-1 accuracy (%): 93.01% and 93.05% | No clear effect |
| include all intermediate timesteps rather than skipping to T3 (ablation) | SNN T1 VGG16 · CIFAR100 | top-1 accuracy (%): 69.92% and 70.15% | No clear effect |
| include all intermediate timesteps rather than skipping to T3 (ablation) | SNN T1 VGG16 · ImageNet | top-1 accuracy (%): 68.98% and 69% | No clear effect |
| split fixed epochs equally across pruning stages (early stop parents) (ablation) | SNN T1 VGG16 · CIFAR10 | top-1 accuracy (%): < 0.3% | No clear effect |
| apply magnitude weight pruning keeping 10% of weights | SNN T1 VGG16 · CIFAR10 | top-1 accuracy (%): 93.05 -> 91.15 | Hurt |
| apply magnitude weight pruning keeping 10% of weights | SNN T1 VGG16 · CIFAR100 | top-1 accuracy (%): 70.15 -> 68.20 | Hurt |
| increase timesteps from 1 to 3 (ablation) | SNN DQN · Cartpole | mean reward: 38.7 -> 52.2 | Improved |
Closest papers by meaning
- Enhancing Training of Spiking Neural Network with Stochastic Latency (AAAI 2024 · similarity 0.91)
- Temporal Flexibility in Spiking Neural Networks: Towards Generalization Across Time Steps and Deployment Friendliness (ICLR 2025 · similarity 0.90)
- Training High Performance Spiking Neural Network by Temporal Model Calibration (ICML 2025 · similarity 0.90)
- TS-SNN: Temporal Shift Module for Spiking Neural Networks (ICML 2025 · similarity 0.90)
- Sparse Spiking Neural Network: Exploiting Heterogeneity in Timescales for Pruning Recurrent SNN (ICLR 2024 · similarity 0.89)
- Enhanced Self-Distillation Framework for Efficient Spiking Neural Network Training (NeurIPS 2025 · similarity 0.89)
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.
