Neuromorphic computing mimics brain efficiency but lacks theoretical guidance. Here, authors develop a computational foundation for processing signals in space and time in spiking neural networks that can outperform standard neural networks in event-based vision.