Calculates the posterior mean projection of data into the latent space.
The posterior mean projection of a point from the target space, t, is the mean of the correspondig posterior distribution induced in the latent space.
means = gtm_pmn(T, X, FI, W, b)
T - data points representing the distribution in the target space. N-by-D
X - data points forming a latent variable sample of the distribution in the latent space. K-by-L
FI - activations of the basis functions when fed X; K-by-(M+1)
W - a matrix of trained weights
b - the trained value for beta
means - the posterior means in latent space. N-by-L