Research map / Label-efficient and robust learning / Knowledge distillation
BKDSNN: Enhancing the Performance of Learning-based Spiking Neural Networks Training with Blurred Knowledge Distillation
Zekai Xu, Kang You, Qinghai Guo, Xiang Wang, Zhezhi He
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 |
|---|---|---|---|
| apply blurred KD on SEW ResNet-50 ImageNet instead of KDSNN/LaSNN | SEW ResNet-50 · ImageNet-1K | top-1 accuracy: 67.78 -> 72.32 | Improved |
| prior layer-wise KD (KDSNN, LaSNN) on ImageNet CNN SNNs | SEW ResNet-18/34/50 · ImageNet-1K | top-1 accuracy: ResNet-50: 67.78 -> 67.72 (KDSNN) | No clear effect |
| add mixed distillation (logits + BKD) rather than either alone (ablation) | SEW ResNet-50 · ImageNet-1K | top-1 accuracy: 71.78 / 71.24 -> 72.32 | Improved |
| replace LaSNN feature loss with BKD loss (ablation) | LaSNN · ImageNet-1K SEW ResNet-18 | top-1 accuracy: 63.31 -> 63.82 | Improved |
| replace KDSNN feature loss with BKD loss (ablation) | KDSNN · ImageNet-1K Sformer-8-384 | top-1 accuracy: 74.62 -> 74.93 | Improved |
| apply KD layer-wise instead of only before last layer (ablation) | BKD · ImageNet-1K | top-1 accuracy | No clear effect |
| remove restoration block G from BKD (ablation) | BKD on Sformer-4-384-400E · CIFAR100 | accuracy: 81.53 -> 77.16 | Hurt |
| increase blurred ratio from 0.00 to 0.60 (ablation) | BKD · CIFAR100 Sformer-4-384-400E | accuracy: 80.44 -> 81.53 | Improved |
| raise logit distillation temperature (ablation) | L_LD on Sformer-4-384-400E · CIFAR10 vs CIFAR100 | accuracy: CIFAR10 96.13 -> 96.31; CIFAR100 81.40 -> 80.88 (tau 1.5 -> 4.0) | Mixed |
| BKD instead of logit KD in terms of firing rate (ablation) | Sformer-4-384-400E · CIFAR10 | firing rate: 5.72 -> 4.67 (e-2) | Improved |
| apply BKDSNN to Sformer-8-768 energy | Sformer-8-768 · ImageNet-1K | energy consumption (mJ): 13.42 -> 13.72 vs Sformer-8-768+CML | Hurt |
| BKDSNN on ResNet-19 CIFAR with limited-knowledge teacher | ResNet-19 / ResNet-20 · CIFAR10/100 | accuracy: ResNet-19 CIFAR100 74.47 -> 74.95 | Improved |
Later papers that compared with this method
3 later accepted papers report BKDSNN as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- A Closer Look at Knowledge Distillation in Spiking Neural Network Training (AAAI 2026 · better on 11 of 11 shared results)
- Bi-Spectrum Distillation: Addressing Spectral Mismatch in ANN-SNN Knowledge Transfer (AAAI 2026 · better on 4 of 4 shared results)
- Many Eyes, One Mind: Temporal Multi-Perspective and Progressive Distillation for Spiking Neural Networks (ICLR 2026 · 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 Separation with Entropy Regularization for Knowledge Distillation in Spiking Neural Networks (CVPR 2025 · similarity 0.91)
- Efficient Logit-based Knowledge Distillation of Deep Spiking Neural Networks for Full-Range Timestep Deployment (ICML 2025 · similarity 0.91)
- Enhanced Self-Distillation Framework for Efficient Spiking Neural Network Training (NeurIPS 2025 · similarity 0.91)
- Synergy Between the Strong and the Weak: Spiking Neural Networks are Inherently Self-Distillers (NeurIPS 2025 · similarity 0.91)
- EnOF-SNN: Training Accurate Spiking Neural Networks via Enhancing the Output Feature (NeurIPS 2024 · similarity 0.90)
- Efficient ANN-Guided Distillation: Aligning Rate-based Features of Spiking Neural Networks through Hybrid Block-wise Replacement (CVPR 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.
