In the literature, they bear the names of

- GMMs: Gaussian mixture models, or
- MoGs: Mixtures of Gaussians.

- The weights
**w[i]**are drawn uniformly from**[0,1]**(and renormalized so that they sum up to 1), - The means
**mu[i]**are also sampled from a uniform distribution in**[0,1]**, - The variance-covariance matrix
**Sigma[i]**(positive definite matrix) is computed as**A A^T**where the elements are drawn from a normal distribution N(0,1) of unit variance centered at zero. The matrix elements are scaled by a factor**epsilon**. A smaller value of**epsilon**separates more the components of the Gaussians.

Frank Nielsen, September 2008.