Research map / Architectures and efficiency / Spiking networks
Temporal Representation Enhancement (TRE): Learning to Forget Dominant Patterns for Enhanced Temporal Spiking Features
Wei Liu, Li Yang, Yufei Wang, Han Xiao, Boyu Cai, Weiming Hu
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 TRE channel suppression to baseline training (ablation) | spiking ResNet-19 · CIFAR-100, T=4 | accuracy (%): 77.75 -> 81.27 | Improved |
| add TRE channel suppression to baseline training (ablation) | spiking ResNet-19 · DVS-CIFAR10, T=10 | accuracy (%): 78.40 -> 83.60 | Improved |
| use TRE instead of TET supervision (ablation) | spiking ResNet-19 · CIFAR-100, T=4 | accuracy (%): 80.16 -> 81.27 | Improved |
| use TRE instead of TET supervision (ablation) | spiking ResNet-19 · DVS-CIFAR10, T=10 | accuracy (%): 83.00 -> 83.60 | Improved |
| use adaptive threshold gating instead of rank-based top-1 gating (ablation) | TRE · CIFAR-100, T=4 | accuracy (%): 80.88 -> 81.27 | Improved |
| suppress more channels with top-k gating (k=1 to 4) (ablation) | TRE · CIFAR-100, T=4 | accuracy (%): 80.88 -> 80.52 | Hurt |
| add TRE to PSN ResNet-34 | PSN · ImageNet, T=4 | top-1 accuracy (%): 70.54 -> 71.60 | Improved |
| add TRE to PSN ResNet-18 | PSN · ImageNet, T=4 | top-1 accuracy (%): 67.63 -> 67.74 | No clear effect |
| add TRE to PSN ResNet-34 | PSN · ImageNet, T=4 | mean off-diagonal cosine similarity: 0.95 -> 0.85 | Improved |
| set adaptive scale lambda to 0.0 with theta0 0.05 (ablation) | TRE · CIFAR-100, T=4 | accuracy (%): 81.27 -> 80.76 | Hurt |
| increase fixed threshold theta0 from 0.05 to 0.06 (ablation) | TRE · CIFAR-100, T=4 | accuracy (%): 81.27 -> 80.85 | Hurt |
| add normalized gating during training (ablation) | TRE · CIFAR-100 | classification score distribution shift | Improved |
Closest papers by meaning
- Towards More Discriminative Feature Learning in SNNs with Temporal-Self-Erasing Supervision (AAAI 2025 · similarity 0.89)
- Temporal Flexibility in Spiking Neural Networks: Towards Generalization Across Time Steps and Deployment Friendliness (ICLR 2025 · similarity 0.86)
- Shrinking Your TimeStep: Towards Low-Latency Neuromorphic Object Recognition with Spiking Neural Networks (AAAI 2024 · similarity 0.86)
- TS-SNN: Temporal Shift Module for Spiking Neural Networks (ICML 2025 · similarity 0.86)
- Enhancing Representation of Spiking Neural Networks via Similarity-Sensitive Contrastive Learning (AAAI 2024 · similarity 0.85)
- Training High Performance Spiking Neural Network by Temporal Model Calibration (ICML 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.
