Research map / Architectures and efficiency / Spiking networks
High-Fidelity ANN-to-SNN Conversion via Closed-Loop CKA Distillation
Bozhou Li, Chubo Liu, Yan Ding, Yufeng Zhang, Zhuo Tang, Kenli Li
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 global distillation loss (ablation) | ResNet-18 task-loss baseline · CIFAR-10 (T=2) | accuracy: 85.08 -> 87.68 | Improved |
| add global distillation loss (ablation) | ResNet-18 task-loss baseline · CIFAR-10 (T=8) | accuracy: 95.10 -> 94.92 | Hurt |
| add global distillation loss (ablation) | ResNet-18 task-loss baseline · CIFAR-10 (T=64) | accuracy: 96.14 -> 96.41 | Improved |
| add CKA local loss (ablation) | ResNet-18 task-loss baseline · CIFAR-10 (T=2) | accuracy: 85.08 -> 88.86 | Improved |
| add CKA local loss (ablation) | ResNet-18 task-loss baseline · CIFAR-10 (T=64) | accuracy: 96.14 -> 96.18 | No clear effect |
| local-only instead of global-only loss (ablation) | ResNet-18 task-loss baseline · CIFAR-10 (T=4) | accuracy: 93.32 -> 93.50 | Improved |
| use beta=0.2 instead of 0.5 (ablation) | dual alignment · CIFAR-10 (T=2) | accuracy: 85.09 -> 90.05 | Improved |
| combine global and local alignment vs baseline (ablation) | ResNet-18 · CIFAR-10 (T=2) | accuracy: 85.08 -> 90.05 | Improved |
| apply closed-loop fine-tuning after SpikingJelly conversion | SpikingJelly · CIFAR-10 ResNet-18 (T=8) | accuracy: 49.05 -> 95.33 | Improved |
| apply closed-loop fine-tuning after SpikingJelly conversion | SpikingJelly · CIFAR-10 ResNet-18 (T=32) | accuracy: 93.90 -> 96.38 | Improved |
| apply closed-loop fine-tuning | Convert-SNN · CIFAR-10 Gaussian noise 0.1 | accuracy: ≈20% -> ≈40% | Improved |
| add refinement stage after conversion | SpikingJelly ann2snn · CIFAR-10 ResNet-18 (T=2) | wall-clock time (s): 245 -> 1003 | Hurt |
Closest papers by meaning
- Inference-Scale Complexity in ANN-SNN Conversion for High-Performance and Low-Power Applications (CVPR 2025 · similarity 0.87)
- Temporal Misalignment in ANN-SNN Conversion and its Mitigation via Probabilistic Spiking Neurons (ICML 2025 · similarity 0.86)
- Efficient ANN-Guided Distillation: Aligning Rate-based Features of Spiking Neural Networks through Hybrid Block-wise Replacement (CVPR 2025 · similarity 0.85)
- One-Time Soft Alignment Enables Resilient Learning without Weight Transport (NeurIPS 2026 · similarity 0.85)
- Efficient ANN-SNN Conversion with Error Compensation Learning (ICML 2025 · similarity 0.85)
- Self-Attentive Spatio-Temporal Calibration for Precise Intermediate Layer Matching in ANN-to-SNN Distillation (AAAI 2025 · similarity 0.85)
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.
