Expressive Power of Recurrent Spiking Neural Networks for Sequence Modeling

Spiking neural networks (SNNs) provide a biologically inspired and potentially efficient framework for processing sequential data, but their expressive power, particularly in the dynamical setting, remains poorly understood. We establish a universality result showing that recurrent SNNs can approximate a broad class of sequence-to-sequence mappings, thereby providing theoretical support for their use in temporal learning and time series processing.