Research map / Architectures and efficiency / Spiking networks
Efficient Training of Spiking Neural Networks with Multi-Parallel Implicit Stream Architecture
Zhigao Cao, Meng Li, Xiashuang Wang, Haoyu Wang, Fan Wang, Youjun Li, Zigang Huang
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 BPTT with implicit equilibrium training (MPIS) on Fashion-MNIST at T=30 | SNN 133K · Fashion-MNIST | accuracy: 89.60% -> 93.14% | Improved |
| replace BPTT with MPIS at T=100 on Fashion-MNIST | SNN 133K · Fashion-MNIST | training memory: 4.8G -> 1.2G | Improved |
| replace BPTT with MPIS on N-MNIST at T=30 | SNN 213K · N-MNIST | training memory: 12.8G -> 3.3G | Improved |
| replace BPTT with MPIS | SNN on N-MNIST · N-MNIST | firing rate (layer 1): 5.06e-2 -> 8.0e-4 | Improved |
| replace IDE-Nets with multi-parallel implicit streams at 30 time steps | CIFAR-10 11.8M · CIFAR-10 | accuracy: 90.37% -> 92.79% | Improved |
| replace IDE-Nets with MPIS at 30 time steps | CIFAR-10 11.8M · CIFAR-10 | training time: 12min10s -> 2min34s | Improved |
| replace IDE-Nets with MPIS at 30 time steps | CIFAR-100 14.8M · CIFAR-100 | accuracy: 70.26% -> 73.19% | Improved |
| scale MPIS parameters to about twice IDE-Net size (ablation) | MPIS-SNNs · CIFAR-100 | accuracy: 73.19% -> 74.40% | Improved |
| use two parallel implicit streams instead of a single stream (ablation) | MPIS-SNNs · N-MNIST, CIFAR-10, CIFAR-100 | convergence speed | Improved |
| increase time steps from 30 to 100 for IDE-Nets (ablation) | IDE-Nets · CIFAR-10 | accuracy: 90.37% -> 90.57% | No clear effect |
| increase time steps from 30 to 100 for MPIS on Fashion-MNIST (ablation) | MPIS · Fashion-MNIST | accuracy: 93.14% -> 93.23% | No clear effect |
Closest papers by meaning
- Efficient Parallel Training Methods for Spiking Neural Networks with Constant Time Complexity (ICML 2025 · similarity 0.90)
- Multiplication-Free Parallelizable Spiking Neurons with Efficient Spatio-Temporal Dynamics (NeurIPS 2025 · similarity 0.89)
- Temporal Flexibility in Spiking Neural Networks: Towards Generalization Across Time Steps and Deployment Friendliness (ICLR 2025 · similarity 0.88)
- Enhanced Self-Distillation Framework for Efficient Spiking Neural Network Training (NeurIPS 2025 · similarity 0.88)
- Faster and Stronger: When ANN-SNN Conversion Meets Parallel Spiking Calculation (ICML 2025 · similarity 0.88)
- Rethinking the Membrane Dynamics and Optimization Objectives of Spiking Neural Networks (NeurIPS 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.
