Output details

Structural Expectation Propagation (SEP): Bayesian Structure Learning for Networks with Latent Variables

<24> Originality: The first time that deterministic approximate Bayesian inference has been applied to structure learning in probabilistic models with hidden variables.

Significance: Expectation propagation is the state-of-the-art algorithm for fast inference of the distribution of variables in graphical models, and this paper extends the technique to learning of the structure of the graphs. Unlike previous techniques Structural Expectation Propagation returns a probability distribution over graphs, reflecting residual uncertainty, rather than a single (greedy) structure.

Rigour: A detailed mathematical derivation of the message-passing algorithm is presented, and the performance is shown to be superior to the previous leading technique.