Research map / Architectures and efficiency / Spiking networks
Bullet Trains: Parallelizing Training of Temporally Precise Spiking Neural Networks
Todd Morrill, Christian-Gernot Pehle, Anthony M. Zador
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 |
|---|---|---|---|
| process input spikes in parallel chunks with associative scans instead of serially | event-based LIF SNN with exact hard reset · SHD training time per batch | speedup: up to 44.27x | Improved |
| increase hidden layer size and batch size (ablation) | parallel SNN · SHD | speedup over serial: 44.27x -> 7.95x | Hurt |
| increase chunk size beyond ~128 (ablation) | parallel SNN · SHD runtime | batch runtime | Mixed |
| use continuous spike times with numerical solvers instead of analytical tau_m=2tau_s | exact-gradient SNN · MNIST 1F350H | accuracy: 97.20 -> 98.04 | Improved |
| use continuous spike times and delays vs discrete-time exact gradients | exact-gradient SNN 2F512HD · SSC | accuracy: 76.10 -> 77.79 | Improved |
| use continuous-time exact gradients instead of surrogate gradient with discrete time | SNN with delays · SSC | accuracy: 80.69 -> 77.79 | Hurt |
| quantize spike times to larger time steps (ablation) | Yin-Yang ITD task · Yin-Yang | test accuracy: near chance (33%) at dt >= 2ms | Hurt |
| use Newton-Raphson instead of Bisection solver (ablation) | spike time solver · convergence | iterations to precision: Bisection needs 20 iterations | Improved |
| use fixed logit temperature multiplier for weighted leaky integrator output (ablation) | output layer logits · classification | stability of softmax | Improved |
Closest papers by meaning
- Temporal Flexibility in Spiking Neural Networks: Towards Generalization Across Time Steps and Deployment Friendliness (ICLR 2025 · similarity 0.91)
- Parallel Training Time-to-First-Spike Spiking Neural Networks (AAAI 2026 · similarity 0.91)
- Efficiently Training Time-to-First-Spike Spiking Neural Networks from Scratch (ICML 2026 · similarity 0.91)
- Efficient Parallel Training Methods for Spiking Neural Networks with Constant Time Complexity (ICML 2025 · similarity 0.90)
- Pseudo-Spiking Neurons: A Noise-Based Training Framework for Heterogeneous-Latency Spiking Neural Networks (AAAI 2026 · similarity 0.90)
- Efficient Logit-based Knowledge Distillation of Deep Spiking Neural Networks for Full-Range Timestep Deployment (ICML 2025 · similarity 0.88)
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.
