Bullet Trains: Parallelizing Training of Temporally Precise Spiking Neural Networks

2026-02-27

AI Summary

This work addresses the challenge of efficiently training event-driven spiking neural networks (SNNs), which is hindered by the sequential nature of hard-reset dynamics and the absence of a differentiable, temporally continuous method for precise spike-time computation. To overcome these limitations, the authors propose a parallel associative scan algorithm that enables, for the first time, simultaneous processing of multiple spikes in native event-based SNNs. Combined with a machine-precision differentiable spike-time solver, this approach preserves the continuous-time hard-reset dynamics while eliminating reliance on time binning. Evaluated on four event-based datasets, the method achieves up to a 44-fold speedup over conventional sequential simulation, demonstrating both superior computational efficiency and high temporal precision in spike timing.

Abstract

Continuous-time, event-native spiking neural networks (SNNs) operate strictly on spike events, treating spike timing and ordering as the representation rather than an artifact of time discretization. This viewpoint aligns with biological computation and with the native resolution of event sensors and neuromorphic processors, while enabling compute and memory that scale with the number of events. However, two challenges hinder practical, end-to-end trainable event-based SNN systems: 1) exact charge--fire--reset dynamics impose inherently sequential processing of input spikes, and 2) precise spike times must be solved without time bins. We address both. First, we use parallel associative scans to consume multiple input spikes at once, yielding up to 44x speedups over sequential simulation while retaining exact hard-reset dynamics. Second, we implement differentiable spike-time solvers that compute spike times to machine precision without discrete-time approximations or restrictive analytic assumptions. We demonstrate the viability of training SNNs using our solutions on four event-based datasets on GPUs.

Problem

• spiking neural networks
• event-based processing
• exact spike timing
• sequential dynamics
• time discretization

Innovation

• spiking neural networks
• event-based processing
• parallel associative scan
• differentiable spike timing
• continuous-time SNNs