Research map / Architectures and efficiency / Spiking networks
Resolving the Timestep Scaling Paradox in Spiking Neural Networks with a Timestep-Scalable Neuron Model
Binghao Ye, Wenjuan Li, Dengfeng Xue, Bing Li, Weiming Hu, Dong Liang, Kun Shang
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 memory reconsolidation (ablation) | LIF neuron · CIFAR-100 (ResNet20, T=10) | accuracy (%): 72.01 -> 74.33 | Improved |
| add memory reconsolidation (ablation) | LIF neuron · CIFAR10-DVS (ResNet20, T=10) | accuracy (%): 79.60 -> 81.50 | Improved |
| add temporal forgetting alone (ablation) | LIF neuron · CIFAR-100 (ResNet20, T=10) | accuracy (%): 72.01 -> 72.85 | Improved |
| add temporal forgetting on top of reconsolidation (ablation) | TS-LIF (R only) · CIFAR-100 (ResNet20, T=10) | accuracy (%): 74.33 -> 75.54 | Improved |
| add temporal forgetting on top of reconsolidation (ablation) | TS-LIF (R only) · CIFAR10-DVS (ResNet20, T=10) | accuracy (%): 81.50 -> 82.30 | Improved |
| add normalization alone (ablation) | LIF neuron · CIFAR-100 (ResNet20, T=10) | accuracy (%): 72.01 -> 69.95 | Hurt |
| add normalization alone (ablation) | LIF neuron · CIFAR10-DVS (ResNet20, T=10) | accuracy (%): 79.60 -> 78.70 | Hurt |
| add normalization on top of forgetting (ablation) | TS-LIF (F only) · CIFAR10-DVS (ResNet20, T=10) | accuracy (%): 80.30 -> 81.20 | Improved |
| add normalization to reconsolidation plus forgetting (ablation) | TS-LIF (R+F) · CIFAR-100 (ResNet20, T=10) | accuracy (%): 75.54 -> 76.28 | Improved |
| replace LIF with TS-LIF at larger timesteps | 3B-Net · Gait (T=90) | accuracy (%): 83.80 -> 89.80 | Improved |
| increase timesteps from 4 to 8 | SpikeYOLO 13.2M · COCO | mAP@50 (%): 51.3 -> 50.1 | Hurt |
| increase timesteps from 4 to 8 | TS-LIF 13.2M · COCO | mAP@50 (%): 51.8 -> 53.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FSTA-SNN — FSTA-SNN:Frequency-Based Spatial-Temporal Attention Module for Spiking Neural Networks (AAAI 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- 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 0.89)
- Rethinking Spiking Neural Networks from an Ensemble Learning Perspective (ICLR 2025 · similarity 0.89)
- Efficient Logit-based Knowledge Distillation of Deep Spiking Neural Networks for Full-Range Timestep Deployment (ICML 2025 · similarity 0.89)
- TS-SNN: Temporal Shift Module for Spiking Neural Networks (ICML 2025 · similarity 0.88)
- Temporal Misalignment in ANN-SNN Conversion and its Mitigation via Probabilistic Spiking Neurons (ICML 2025 · similarity 0.88)
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.
