Sparse Axonal and Dendritic Delays Enable Competitive SNNs for Keyword Classification

TL;DR

The paper investigates learning axonal and dendritic delays in deep feedforward SNNs to improve temporal processing with reduced memory and buffering. Using LIF neurons and a DCLS-based delay learning approach, the authors demonstrate competitive accuracy on keyword classification benchmarks (GSC and SSC) with significant efficiency gains over synaptic delays. Axonal delays offer the best trade-offs in buffering and performance, and both delay types remain robust under substantial delay sparsity, broadening their applicability to resource-constrained neuromorphic hardware. The study provides a practical, scalable mechanism for temporal representation in SNNs and contributes a transparent framework for comparing delay mechanisms.

Abstract

Training transmission delays in spiking neural networks (SNNs) has been shown to substantially improve their performance on complex temporal tasks. In this work, we show that learning either axonal or dendritic delays enables deep feedforward SNNs composed of leaky integrate-and-fire (LIF) neurons to reach accuracy comparable to existing synaptic delay learning approaches, while significantly reducing memory and computational overhead. SNN models with either axonal or dendritic delays achieve up to $95.58\%$ on the Google Speech Command (GSC) and $80.97\%$ on the Spiking Speech Command (SSC) datasets, matching or exceeding prior methods based on synaptic delays or more complex neuron models. By adjusting the delay parameters, we obtain improved performance for synaptic delay learning baselines, strengthening the comparison. We find that axonal delays offer the most favorable trade-off, combining lower buffering requirements with slightly higher accuracy than dendritic delays. We further show that the performance of axonal and dendritic delay models is largely preserved under strong delay sparsity, with as few as $20\%$ of delays remaining active, further reducing buffering requirements. Overall, our results indicate that learnable axonal and dendritic delays provide a resource-efficient and effective mechanism for temporal representation in SNNs. Code will be made available publicly upon acceptance. Code is available at https://github.com/YounesBouhadjar/AxDenSynDelaySNN

Figures (3)

• Figure 1: Test accuracy as a function of buffer size for (A) SSC and (B) GSC benchmarks. Results compare DCLS-based models with axonal, dendritic, and synaptic delays (ours) against baseline spiking neuron models, including SiLIF, cAdLIF, d-cAdLIF, and SE-AdLIF. The selected DCLS-based models correspond to the best-performing configurations reported in \ref{['tab:performance_models_comparison']}. Buffer size is shown on a logarithmic scale. Points denote mean test accuracy across runs, with error bars indicating standard deviation.
• Figure 2: Effect of delay range and sparsity on classification accuracy. (A--B) Test accuracy as a function of the maximum allowed delay $d_{\max}$ for synaptic, axonal, and dendritic delay models, evaluated on GSC (A) and SSC (B). (C--D) Test accuracy of axonal-delay models as a function of delay sparsity $\eta$ for different levels of weight sparsity $\kappa$, evaluated on GSC (C) and SSC (D).
• Figure 3: Effect of firing-rate regularization on axonal and dendritic delay models. Test accuracy (black) and synaptic operations (SOP, gray) as a function of the maximum allowed firing rate ($\alpha_\text{max}$) under activity regularization. The minimum firing rate ($\alpha_\text{min}$) is set to 0.001 and the regularization factor ($r$) to 0.5. (A) Axonal delays. (B) Dendritic delays.