高速列车：时间精确脉冲神经网络训练的并行化

Continuous-time, event-driven spiking neural networks (SNNs) operate strictly on spike events, treating spike timing and ordering as representations rather than artifacts of time discretization. This perspective aligns with biological computation and the native resolution of event sensors and neuromorphic processors, enabling compute and memory usage that scales 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 resolved without time bins. To address these issues, parallel associative scans are employed to consume multiple input spikes simultaneously, enhancing processing efficiency.