Covariance Structures in Gaussian Mixtures
Constraining each component's Sigma — full (any ellipsoid, O(D^2) parameters), diagonal (axis-aligned), tied (one shared Sigma), or spherical (a single variance) — trades geometric flexibility against parameter count, overfitting risk, and per-iteration cost.
This Concept is waiting for its first lesson!
Constraining each component's Sigma — full (any ellipsoid, O(D^2) parameters), diagonal (axis-aligned), tied (one shared Sigma), or spherical (a single variance) — trades geometric flexibility against parameter count, overfitting risk, and per-iteration cost.
Are you a teacher? Sign in to start contributing.
Sign In