Mixtures of Gaussians as a Proxy in Hybrid Bayesian Networks

In this paper we explore the use of mixtures of Gaussians as a proxy for mixtures of truncated basis functions in hybrid Bayesian networks. The idea is to use mixtures of Gaussians during the learning process, and move to mixtures of truncated basis functions for carrying out probabilistic inference. This would bridge the gap between efficient inference and learning in hybrid Bayesian networks, specially in scenarios where data comes in streams and models need to be continuously updated.