Research map / Architectures and efficiency / Spiking networks
Sparse Spiking Gradient Descent
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 sparse backpropagation through active neurons instead of dense backpropagation | Three-layer fully connected SNN · Fashion-MNIST | test accuracy: 82.2% | No clear effect |
| Use sparse backpropagation through active neurons instead of dense backpropagation | Three-layer fully connected SNN · Neuromorphic-MNIST | test accuracy: 92.7% | No clear effect |
| Use sparse backpropagation through active neurons instead of dense backpropagation | Three-layer fully connected SNN · SHD | test accuracy: 77.5% | No clear effect |
| Use sparse backpropagation through active neurons instead of dense backpropagation | Three-layer fully connected SNN · Fashion-MNIST | backward speedup: 40x | Helped |
| Use sparse backpropagation through active neurons instead of dense backpropagation | Three-layer fully connected SNN · Fashion-MNIST | GPU memory saved: 35% | Helped |
| Use sparse backpropagation through active neurons instead of dense backpropagation | Six-layer fully connected SNN · Fashion-MNIST | backward speedup: 15x | Helped |
| Use sparse backpropagation through active neurons instead of dense backpropagation | Six-layer fully connected SNN · Fashion-MNIST | memory saved: 25% | Helped |
| Approximate spike gradients sparsely using active neurons | Convolutional SNN · Fashion-MNIST | test accuracy: 86.7% vs 86.9% | No clear effect |
| Increase the backpropagation threshold to B_th=0.95 (ablation) | Sparse backpropagation · SHD | test accuracy: B_th=0.95 | No clear effect |
| Increase the backpropagation threshold to B_th=0.95 (ablation) | Sparse backpropagation · SHD | backward speedup: nearly 150x | Helped |
| Increase the backpropagation threshold to B_th=0.95 (ablation) | Sparse backpropagation · SHD | backward memory saved: 85% | Helped |
| Increase the backpropagation threshold toward 1 until no gradients can propagate (ablation) | Sparse backpropagation · SHD | gradient activity | Hurt |
Closest papers by meaning
- Improving the Sparse Structure Learning of Spiking Neural Networks from the View of Compression Efficiency (ICLR 2025 · similarity 0.92)
- SparseProp: Efficient Event-Based Simulation and Training of Sparse Recurrent Spiking Neural Networks (NeurIPS 2023 · similarity 0.92)
- SpikeGrad: An ANN-equivalent Computation Model for Implementing Backpropagation with Spikes (ICLR 2020 · similarity 0.92)
- Sparse Spiking Neural Network: Exploiting Heterogeneity in Timescales for Pruning Recurrent SNN (ICLR 2024 · similarity 0.91)
- Enhanced Self-Distillation Framework for Efficient Spiking Neural Network Training (NeurIPS 2025 · similarity 0.90)
- Advancing Training Efficiency of Deep Spiking Neural Networks through Rate-based Backpropagation (NeurIPS 2024 · similarity 0.90)
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.