Research map / Architectures and efficiency / Spiking networks
Adaptive Smoothing Gradient Learning for Spiking Neural Networks
Ziming Wang, Runhao Jiang, Shuang Lian, Rui Yan, Huajin Tang
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 |
|---|---|---|---|
| adaptive (learnable) smoothing width with dual-mode forwarding instead of constant surrogate width (ablation) | SNN trained on CIFAR-10/CIFAR-100 (ResNet-19/ResNet-18) · CIFAR-10 and CIFAR-100 ablation | accuracy (%): CIFAR-10 width 10.0: 30.85 to 93.53; CIFAR-100 width 1.0: 65.19 to 76.09 | Improved |
| adaptive width learning (vs fixed width) under full-spike and p=0.8 noise (ablation) | ResNet-19 on CIFAR-10, 2 steps · CIFAR-10 ablation (Table 3) | accuracy (%): up to about +1 point at p=1.0 and p=0.8 across widths | Improved |
| dual-mode switching (noise probability p=0.8 instead of full spikes p=1.0) (ablation) | ResNet-19 on CIFAR-10 with adaptive width · CIFAR-10, 2 steps | accuracy (%): 92.94 to 93.32 | Improved |
| noise probability p set too low or too high (fixed p sweep) (ablation) | ResNet-19 on CIFAR-10 · CIFAR-10 noise probability analysis | test accuracy | Mixed |
| ASGL training instead of surrogate gradient learning on SHD recurrent SNN | 700-240-20 recurrent SNN · SHD | accuracy (%): 86.9 vs 84.8 for EventProp (best baseline listed) | Improved |
| ASGL training on DVS-CIFAR10 with VGGSNN | VGGSNN · DVS-CIFAR10 | accuracy: 84.50 vs 83.17 for TET | Improved |
| ASGL training on Tiny-ImageNet with VGG-13 at 8 time steps | VGG-13 · Tiny-ImageNet | accuracy (%): 56.81 vs 55.37 for offline LTL | Improved |
Closest papers by meaning
- Take A Shortcut Back: Mitigating the Gradient Vanishing for Training Spiking Neural Networks (NeurIPS 2024 · similarity 0.90)
- Rethinking Spiking Neural Networks from an Ensemble Learning Perspective (ICLR 2025 · similarity 0.90)
- Adaptive Surrogate Gradients for Sequential Reinforcement Learning in Spiking Neural Networks (NeurIPS 2025 · similarity 0.90)
- Temporal Flexibility in Spiking Neural Networks: Towards Generalization Across Time Steps and Deployment Friendliness (ICLR 2025 · similarity 0.90)
- NDOT: Neuronal Dynamics-based Online Training for Spiking Neural Networks (ICML 2024 · similarity 0.89)
- Rethinking the Membrane Dynamics and Optimization Objectives of Spiking Neural Networks (NeurIPS 2024 · 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.
