Fusion GMBP

Fusion GMBP is a software family for Gaussian Mixture Belief Propagation: exact and approximate inference over probabilistic graphical models with Gaussian-mixture factors. It connects model learning, inference, mixture reduction, and application integration.

Components

hybn learns hybrid Bayesian network structure and parameters from tabular data containing discrete, continuous, or mixed variables.
gmbp provides mixture operations and belief-propagation schedules: exact on trees and forests, approximate on cyclic graphs.
gmred reduces Gaussian and hybrid mixtures under stated preservation objectives, then measures the reduction error.
gmkit connects learning, compiled inference, and streaming evidence behind an application-facing uncertainty layer.

Public reference

The KIT GitLab gmbp project describes the public inference core and credits implementation by Alper Saritas under the supervision of Luisa Hoffmann. The MFI 2026 publication covers related Gaussian mixture reduction research.