Skip to content

multivariate normal distribution

3-dim projections of 4-dim MVN

Covariance matrix of a cut-off Multivariate Normal Distribution – II – integrals over volume and surface of an n-dimensional sphere

In the 1st post of this series, we have posed the following problem: Take the probability density of a Multivariate Normal Distribution [MVN], but set it to zero at Mahalanobis distances bigger than a finite Mahalanobis distance D. Take a respective volume region enclosed by a contour surface of constant probability density, for a Mahalanobis distance dm ≤ D. We… Read More »Covariance matrix of a cut-off Multivariate Normal Distribution – II – integrals over volume and surface of an n-dimensional sphere

Cut-off BVN limited to an ellipsoidal core

Covariance matrix of a cut-off Multivariate Normal Distribution – I – intricate integrals with exponentials over the volumes and surfaces of n-dimensional ellipsoids?

In Machine Learning and statistics one sometimes has to work with a data sample whose underlying probability distribution approximates a Multivariate Normal Distribution [MVN] – but only within the ellipsoidal region of a central core. The core’s surface is assumed to reflect a contour surface of the MVN and would therefore be given by a finite Mahalanobis distance D. I.e.,… Read More »Covariance matrix of a cut-off Multivariate Normal Distribution – I – intricate integrals with exponentials over the volumes and surfaces of n-dimensional ellipsoids?

Iterative method to compute the covariance-matrix of normal MVN-like inner cores of multivariate distributions with strongly asymmetric outer layers – I

In other posts in this blog (see [1] to [3]), I have discussed multiple methods to calculate and construct confidence ellipses of “Bivariate Normal Distributions” [BVNs]. BVNs are the marginal distributions of “Multivariate Normal Distributions” [MVNs] in e.g. n dimensions ( n > 2). Therefore, two-dimensional confidence ellipses appear as projections of n-dimensional concentric confidence ellipsoids of MVNs onto (2-dim) coordinate planes. The properties of the confidence ellipsoids, which also give us contours of the probability density, are defined by the variance-covariance matrix Σ of the MVN. This post discusses a method to compute the confidence ellipsoids and ellipses for an inner MVN-like core of an otherwise largely asymmetric distribution, which in its overall shape and structure deviates strongly from a MVN.

Read More »Iterative method to compute the covariance-matrix of normal MVN-like inner cores of multivariate distributions with strongly asymmetric outer layers – I
CAE generated face on background of a MND

Latent space distribution of a CAE for face images – I – unenforced Multivariate Normal Distributions

The analysis of face images by a trained Autoencoder and the generation of face images from statistical vectors is a classical task in Machine Learning. In this post series I want to clarify the properties of vector distributions for face images generated by a trained standard Convolutional Autoencoder [CAE] in its latent space. The dataset primarily used is the CelebA… Read More »Latent space distribution of a CAE for face images – I – unenforced Multivariate Normal Distributions