Affinity Propagation: Clustering by Message Passing

Affinity Propagation is an unsupervised clustering algorithm that discovers both its exemplars and its cluster count on its own, purely by exchanging "responsibility" and "availability" messages between every pair of points until the network settles. This simulator renders an unlabeled 3D point cloud and runs the real responsibility/availability update equations live in your browser, redrawing which points have become exemplars and which points now belong to them every frame. Live readouts track the iteration count, the number of exemplars the network has settled on, the size of the largest message change this step, and whether the assignment has converged.