spiking neural network

A spiking neural network is a kind of artificial neural network whose units talk to each other the way real neurons do: by firing short, all-or-nothing electrical blips called spikes. Instead of every unit constantly passing along a smooth number, each one stays quiet until enough signal builds up inside it, then suddenly pops off a single spike and falls silent again. Picture a row of people who only shout the instant a bucket they are filling overflows, then start filling from empty once more. The message lives not in how loud each shout is, but in when the shouts happen and how often.

This timing-based style makes spiking networks closer to the brain than the more common artificial networks, where every unit outputs a steady value on every step. Because a spiking unit does nothing until it fires, the network can sit mostly idle and only spend energy on the rare moments when spikes actually occur. That sparseness is why researchers pair spiking networks with neuromorphic chips, special hardware built to run them cheaply. They are studied both as practical low-power machine-learning systems and as simplified models that help scientists understand how living brains might compute using spikes.