attractor network
An attractor network is a way of describing a group of brain cells that are richly wired back into each other, so that their activity naturally rolls toward, and then settles into, a few stable patterns — a bit like a marble released on a bumpy landscape that always rolls down into one of a handful of dips and rests there. Each dip, called an attractor, is one steady state the network can hold, and it stands for something the brain wants to keep firm: a remembered face, a chosen direction, a decision that has been made. Because the cells feed their signals back to one another in loops (this looping wiring is called recurrence), the pattern can keep itself alive on its own, holding the answer steady even after the thing that triggered it is gone.
What makes this so useful is that the network cleans up and completes messy input. Nudge it part of the way toward a dip — give it a blurry hint or only half of a memory — and the recurrent feedback pulls the whole activity pattern the rest of the way down into the nearest stable state, recovering the complete answer; this is how a single clue can bring back a full memory. The same idea explains decisions: as evidence arrives, the network drifts until it tips into one dip rather than another, and that final resting pattern is the choice. Small disturbances or noise get smoothed out because the activity slides back into the dip, which is why these networks can hold information reliably for seconds at a time and recover gracefully from interference.
Researchers use attractor networks both as a theory of how real circuits work and as a tool they can simulate on a computer. The picture fits several brain systems: cells in the hippocampus and cortex are thought to store memories as attractors, head-direction cells appear to settle onto a ring of states that act like an internal compass, and circuits that keep a decision or a held thought in mind seem to lean on the same self-sustaining loops. The model also predicts characteristic failures — for example, if two stored patterns sit too close together, the marble can roll into the wrong dip, blending or confusing memories.
The stable patterns can be single steady states (point attractors) or whole loops and surfaces of states, such as the ring that lets head-direction cells act as a compass.