Research map / Architectures and efficiency / Spiking networks
DS-ATGO: Dual-Stage Synergistic Learning via Forward Adaptive Threshold and Backward Gradient Optimization for Spiking Neural Networks
Jiaqiang Jiang, Wenfeng Xu, Jing Fan, Rui Yan
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 adaptive threshold (AT) (ablation) | Vanilla-SNN · CIFAR10 / CIFAR10-DVS | accuracy | Improved |
| add threshold-driven gradient optimization (TGO) (ablation) | Vanilla-SNN · CIFAR10 / CIFAR10-DVS | accuracy | Improved |
| combine AT and TGO (ablation) | Vanilla-SNN · CIFAR10 | accuracy (%): +1.71% | Improved |
| combine AT and TGO (ablation) | Vanilla-SNN · CIFAR10-DVS | accuracy (%): +2.30% | Improved |
| use true MPD distribution instead of estimated normal distribution (ablation) | DS-ATGO · CIFAR100 / CIFAR10-DVS | accuracy | Improved |
| add moving average of thresholds for inference (ablation) | DS-ATGO · CIFAR100 / CIFAR10-DVS | accuracy | Improved |
| use AT to balance firing rates | DS-ATGO · CIFAR10 ResNet-19 | firing rate stability: 10.66% avg (Vanilla) -> 15±1.62% | Improved |
| learned thresholds (DIET-SNN, LTMD) vs fixed threshold | DIET-SNN, LTMD · CIFAR10 ResNet-19 | firing rate stability | Mixed |
| use TGO instead of fixed SG | DS-ATGO · CIFAR100 ResNet-19 | gradient-available proportion (%): 15.13% -> over 38.53% | Improved |
| adjust SG width via threshold (TGO) | DS-ATGO · CIFAR100 ResNet-19 | loss landscape | Improved |
| combine both adaptive threshold and SG optimization vs single-path methods | DS-ATGO · CIFAR100 | accuracy (%): 80.59 vs 78.84 (MPD-AGL) | Improved |
| use DS-ATGO instead of SNN-ViT on CIFAR100 with fewer timesteps | DS-ATGO · CIFAR100 | accuracy (%): 80.01 (T=4) -> 80.59±0.17 (T=2) | 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.
- QP-SNN — QP-SNN: Quantized and Pruned Spiking Neural Networks (ICLR 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
- Adaptive Surrogate Gradients for Sequential Reinforcement Learning in Spiking Neural Networks (NeurIPS 2025 · similarity 0.91)
- Rethinking the Membrane Dynamics and Optimization Objectives of Spiking Neural Networks (NeurIPS 2024 · similarity 0.90)
- Take A Shortcut Back: Mitigating the Gradient Vanishing for Training Spiking Neural Networks (NeurIPS 2024 · similarity 0.90)
- NDOT: Neuronal Dynamics-based Online Training for Spiking Neural Networks (ICML 2024 · similarity 0.90)
- The Dormant Spiking Neuron: A State-Driven Mechanism for Efficient Spiking Neural Networks (NeurIPS 2026 · similarity 0.89)
- P-SPIKESSM: HARNESSING PROBABILISTIC SPIKING STATE SPACE MODELS FOR LONG-RANGE DEPENDENCY TASKS (ICLR 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.
