202412121424
Status: #idea
Tags:
State: #nascient

Gaussian Mixture Models

We try to find:

1. The number of gaussians
2. Their location (mean)
3. Their width (covariance)

We assign a label based on the generator which maximizes the posterior probability.
We might have overlapping labels where points belong to different clusters to varying degrees, so this is a soft clustering method as opposed to something like K-Means Clustering or Spectral Clustering where there is no overlap.

It uses the Expectation-Maximization Algorithm to find the parameters.

Can cope of different sizes much better than K-Means Clustering

Relevant Links