Probability densities are calculated. According to an example embodiment, a Gaussian probability density is computed for a set of data representing a multivariate Gaussian distribution (Gaussian) by matching moments of truncated distributions. A distribution is projected onto dimensions of the set of data to generate cavity-local Gaussians. For the set of data, site-local univariate Gaussians are selected so that the moments match cavity-local univariate Gaussians (the moments of the product of the cavity-local Gaussians and the selected site-local univariate Gaussians, match the moments of the product of the cavity-local Gaussians and the cavity-local truncations). The selected site-local univariate Gaussians are aggregated globally to form a new global Gaussian.
展开▼