Soft Versus Hard Cluster Assignment
K-means assigns each point wholly to its nearest centroid, which is the zero-covariance, equal-weight limit of a spherical GMM; GMM responsibilities instead spread membership fractionally, so overlapping or elliptical clusters and per-cluster uncertainty are representable.
This Concept is waiting for its first lesson!
K-means assigns each point wholly to its nearest centroid, which is the zero-covariance, equal-weight limit of a spherical GMM; GMM responsibilities instead spread membership fractionally, so overlapping or elliptical clusters and per-cluster uncertainty are representable.
Are you a teacher? Sign in to start contributing.
Sign In