稀疏轴突和树突延迟使竞争性脉冲神经网络在关键词分类中表现出色

Training transmission delays in spiking neural networks (SNNs) has been shown to significantly enhance their performance on complex temporal tasks. Learning either axonal or dendritic delays allows deep feedforward SNNs composed of leaky integrate-and-fire (LIF) neurons to achieve accuracy comparable to existing synaptic delay learning approaches while substantially reducing memory and computational overhead. SNN models with either axonal or dendritic delays achieve up to 95.58% accuracy on the Google Speech Command (GSC) dataset and 80.97% on the Spiking Speech Command (SSC) dataset, matching or exceeding prior methods based on synaptic delays or more complex neuron models. By adjusting the delay parameters, further performance improvements are obtained.