Sparse Axonal and Dendritic Delays Enable Competitive SNNs for Keyword Classification

Researchers have developed a new method for training Spiking Neural Networks (SNNs) that dramatically cuts computational costs while matching top performance. By learning sparse 'axonal or dendritic delays' instead of complex synaptic delays, the models achieved up to 95.58% accuracy on keyword classification tasks. Crucially, performance was largely preserved even when 80% of the delays were removed, slashing memory and processing overhead. This makes powerful, real-time AI on low-power devices much more feasible.