RBN

rBN is an acronym used in several disciplines, and its meaning depends on the context. In machine learning and probabilistic modeling, the two most common interpretations are recurrent Bayesian networks and randomized Bayesian networks.

A recurrent Bayesian network refers to a dynamic probabilistic graphical model that extends conventional Bayesian networks to handle temporal sequences. In these models, the network is organized into time slices with intra-slice and inter-slice dependencies, allowing the representation of how hidden and observed variables evolve over time. Inference and learning typically employ methods related to dynamic Bayesian networks, such as belief propagation through time or EM-based parameter estimation. Recurrent Bayesian networks are used for time-series forecasting, speech and activity recognition, and other tasks involving sequential data. They are related in spirit to state-space models and to recurrent neural networks, though rBNs retain explicit probabilistic graphical structure and interpretability.

A randomized Bayesian network (or a Bayesian network with randomized components) describes a network where the structure and/or parameters are treated as random variables. This perspective is common in model evaluation, Bayesian structure learning, and simulation studies, where one samples networks from prior distributions to assess robustness, average performance across model families, or to perform Bayesian model averaging. Such approaches help quantify uncertainty in the learned structure and parameters.

Because rBN can denote multiple concepts, the intended meaning typically depends on the field and specific usage. If a context is provided, a more precise definition can be given.