Research map / Architectures and efficiency / Spiking networks
DCT-SNN: Using DCT To Distribute Spatial Information Over Time for Low-Latency Spiking Neural Networks
Isha Garg, Sayeed Shafayet Chowdhury, Kaushik Roy
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 |
|---|---|---|---|
| use orthonormal transform instead of random transform for input encoding (ablation) | DCT-SNN encoding, VGG5 · CIFAR-10 | accuracy (%): 64.7 -> 83.3 | Improved |
| use DCT matrix instead of random unranked orthonormal matrix (ablation) | DCT-SNN encoding, VGG5 · CIFAR-10 | accuracy (%): 83.3 -> 83.5 | Improved |
| use data-dependent PCA bases instead of fixed DCT (ablation) | DCT-SNN encoding, VGG5 · CIFAR-10 | accuracy (%): 83.5 -> 83.8 | No clear effect |
| reduce DCT block size from 16 to 4 (ablation) | DCT-SNN · CIFAR-10 | accuracy (%) | Improved |
| use block size 2 for DCT encoding (ablation) | DCT-SNN · CIFAR-10 | accuracy (%) | Hurt |
| add overlapping blocks with stride 2 to DCT-4 (ablation) | DCT-SNN · CIFAR-10 | accuracy (%) | Improved |
| increase timesteps from 32 to 48 (three cycles) (ablation) | DCT-SNN DCT-4 · CIFAR-10 | accuracy (%): 88.41 -> 89.94 | Improved |
| use DCT-SNN encoding instead of Poisson pixel encoding | SNN VGG9 · CIFAR-10 | accuracy (%): 88.9 (100) -> 89.94 (48) | Mixed |
| use DCT-SNN encoding instead of Poisson-encoded DCT coefficients | SNN VGG9 · CIFAR-10 | accuracy (%): 88.2 (100) -> 89.94 (48) | Improved |
| use DCT-SNN encoding instead of Poisson-encoded DCT coefficients | SNN VGG11 · CIFAR-100 | accuracy (%): 65.1 (125) -> 68.3 (48) | Improved |
| initialize SNN from pretrained ANN instead of training from scratch (ablation) | DCT-SNN VGG9 · CIFAR-10 | accuracy (%): 84.9 -> 89.94 | Improved |
| include all 16 frequencies at inference instead of only first 4 (ablation) | DCT-SNN VGG9 · CIFAR-10 | accuracy (%): 73.9 -> 88.6 | Improved |
Closest papers by meaning
- Efficient Logit-based Knowledge Distillation of Deep Spiking Neural Networks for Full-Range Timestep Deployment (ICML 2025 · similarity 0.88)
- Spiking Vision Transformer with Saccadic Attention (ICLR 2025 · similarity 0.88)
- STAA-SNN: Spatial-Temporal Attention Aggregator for Spiking Neural Networks (CVPR 2025 · similarity 0.88)
- Enhancing Training of Spiking Neural Network with Stochastic Latency (AAAI 2024 · similarity 0.88)
- Inference-Scale Complexity in ANN-SNN Conversion for High-Performance and Low-Power Applications (CVPR 2025 · similarity 0.88)
- LM-HT SNN: Enhancing the Performance of SNN to ANN Counterpart through Learnable Multi-hierarchical Threshold Model (NeurIPS 2024 · similarity 0.87)
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.
