Do detailed derivations of EM algorithm for GMM(Gaussian mixture model), in the case of arbitrary covariance matrices.
Gaussian mixture model is a family of distributions whose pdf is in the following form : K gmm(x) = p(x) = Σπ.(x|μ., Σκ), (1) k=1 where N(μ, E) denotes the Gaussian pdf with mean and covariance matrix Σ, and {₁,..., K} are mixing coefficients satisfying K Tk=p(y=k), TK = 1₁ Tk 20, k={1,..., K}. 2-1 (2) k=1