Sparse / spike-based coding

Neuromorphic efficiency rests on sparsity: representing and transmitting information with few active events, so most of the system stays quiet. Spike-based codes carry information in the identity and precise timing of events rather than in dense continuous values, which suits event-driven hardware and AER buses. Encodings range from rate and population codes to temporal codes such as time-to-first-spike or rank order, and from binary to few-bit graded spikes.

Choosing an encoding trades accuracy, latency, and energy: temporal codes can be fast and sparse but fragile to jitter and noise, while rate-like codes are robust but denser and slower. Enforcing sparsity in a trained SNN (through regularisation or thresholds) is often what converts a nominal architecture into an actually low-energy one.